5 * One Ring to rule them all, One Ring to find them
7 * [p.v of _The Lord of the Rings_, opening poem]
8 * [p.50 of _The Lord of the Rings_, I/iii: "The Shadow of the Past"]
9 * [p.254 of _The Lord of the Rings_, II/ii: "The Council of Elrond"]
12 /* This file contains functions for executing a regular expression. See
13 * also regcomp.c which funnily enough, contains functions for compiling
14 * a regular expression.
16 * This file is also copied at build time to ext/re/re_exec.c, where
17 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
18 * This causes the main functions to be compiled under new names and with
19 * debugging support added, which makes "use re 'debug'" work.
22 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
23 * confused with the original package (see point 3 below). Thanks, Henry!
26 /* Additional note: this code is very heavily munged from Henry's version
27 * in places. In some spots I've traded clarity for efficiency, so don't
28 * blame Henry for some of the lack of readability.
31 /* The names of the functions have been changed from regcomp and
32 * regexec to pregcomp and pregexec in order to avoid conflicts
33 * with the POSIX routines of the same names.
36 #ifdef PERL_EXT_RE_BUILD
41 * pregcomp and pregexec -- regsub and regerror are not used in perl
43 * Copyright (c) 1986 by University of Toronto.
44 * Written by Henry Spencer. Not derived from licensed software.
46 * Permission is granted to anyone to use this software for any
47 * purpose on any computer system, and to redistribute it freely,
48 * subject to the following restrictions:
50 * 1. The author is not responsible for the consequences of use of
51 * this software, no matter how awful, even if they arise
54 * 2. The origin of this software must not be misrepresented, either
55 * by explicit claim or by omission.
57 * 3. Altered versions must be plainly marked as such, and must not
58 * be misrepresented as being the original software.
60 **** Alterations to Henry's code are...
62 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
63 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
64 **** by Larry Wall and others
66 **** You may distribute under the terms of either the GNU General Public
67 **** License or the Artistic License, as specified in the README file.
69 * Beware that some of this code is subtly aware of the way operator
70 * precedence is structured in regular expressions. Serious changes in
71 * regular-expression syntax might require a total rethink.
74 #define PERL_IN_REGEXEC_C
77 #ifdef PERL_IN_XSUB_RE
83 #define RF_tainted 1 /* tainted information used? e.g. locale */
84 #define RF_warned 2 /* warned about big count? */
86 #define RF_utf8 8 /* Pattern contains multibyte chars? */
88 #define UTF_PATTERN ((PL_reg_flags & RF_utf8) != 0)
90 #define RS_init 1 /* eval environment created */
91 #define RS_set 2 /* replsv value is set */
97 /* Valid for non-utf8 strings, non-ANYOFV nodes only: avoids the reginclass
98 * call if there are no complications: i.e., if everything matchable is
99 * straight forward in the bitmap */
100 #define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,0,0) \
101 : ANYOF_BITMAP_TEST(p,*(c)))
107 #define CHR_SVLEN(sv) (utf8_target ? sv_len_utf8(sv) : SvCUR(sv))
108 #define CHR_DIST(a,b) (PL_reg_match_utf8 ? utf8_distance(a,b) : a - b)
110 #define HOPc(pos,off) \
111 (char *)(PL_reg_match_utf8 \
112 ? reghop3((U8*)pos, off, (U8*)(off >= 0 ? PL_regeol : PL_bostr)) \
114 #define HOPBACKc(pos, off) \
115 (char*)(PL_reg_match_utf8\
116 ? reghopmaybe3((U8*)pos, -off, (U8*)PL_bostr) \
117 : (pos - off >= PL_bostr) \
121 #define HOP3(pos,off,lim) (PL_reg_match_utf8 ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
122 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
124 /* these are unrolled below in the CCC_TRY_XXX defined */
126 /* Often 'str' is a hard-coded utf8 string instead of utfebcdic. so just
127 * skip the check on EBCDIC platforms */
128 # define LOAD_UTF8_CHARCLASS(class,str) LOAD_UTF8_CHARCLASS_NO_CHECK(class)
130 # define LOAD_UTF8_CHARCLASS(class,str) STMT_START { \
131 if (!CAT2(PL_utf8_,class)) { \
133 ENTER; save_re_context(); \
134 ok=CAT2(is_utf8_,class)((const U8*)str); \
135 assert(ok); assert(CAT2(PL_utf8_,class)); LEAVE; } } STMT_END
138 /* Doesn't do an assert to verify that is correct */
139 #define LOAD_UTF8_CHARCLASS_NO_CHECK(class) STMT_START { \
140 if (!CAT2(PL_utf8_,class)) { \
141 bool throw_away PERL_UNUSED_DECL; \
142 ENTER; save_re_context(); \
143 throw_away = CAT2(is_utf8_,class)((const U8*)" "); \
146 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS(alnum,"a")
147 #define LOAD_UTF8_CHARCLASS_DIGIT() LOAD_UTF8_CHARCLASS(digit,"0")
148 #define LOAD_UTF8_CHARCLASS_SPACE() LOAD_UTF8_CHARCLASS(space," ")
150 #define LOAD_UTF8_CHARCLASS_GCB() /* Grapheme cluster boundaries */ \
151 LOAD_UTF8_CHARCLASS(X_begin, " "); \
152 LOAD_UTF8_CHARCLASS(X_non_hangul, "A"); \
153 /* These are utf8 constants, and not utf-ebcdic constants, so the \
154 * assert should likely and hopefully fail on an EBCDIC machine */ \
155 LOAD_UTF8_CHARCLASS(X_extend, "\xcc\x80"); /* U+0300 */ \
157 /* No asserts are done for these, in case called on an early \
158 * Unicode version in which they map to nothing */ \
159 LOAD_UTF8_CHARCLASS_NO_CHECK(X_prepend);/* U+0E40 "\xe0\xb9\x80" */ \
160 LOAD_UTF8_CHARCLASS_NO_CHECK(X_L); /* U+1100 "\xe1\x84\x80" */ \
161 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LV); /* U+AC00 "\xea\xb0\x80" */ \
162 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LVT); /* U+AC01 "\xea\xb0\x81" */ \
163 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LV_LVT_V);/* U+AC01 "\xea\xb0\x81" */\
164 LOAD_UTF8_CHARCLASS_NO_CHECK(X_T); /* U+11A8 "\xe1\x86\xa8" */ \
165 LOAD_UTF8_CHARCLASS_NO_CHECK(X_V) /* U+1160 "\xe1\x85\xa0" */
167 #define PLACEHOLDER /* Something for the preprocessor to grab onto */
169 /* The actual code for CCC_TRY, which uses several variables from the routine
170 * it's callable from. It is designed to be the bulk of a case statement.
171 * FUNC is the macro or function to call on non-utf8 targets that indicate if
172 * nextchr matches the class.
173 * UTF8_TEST is the whole test string to use for utf8 targets
174 * LOAD is what to use to test, and if not present to load in the swash for the
176 * POS_OR_NEG is either empty or ! to complement the results of FUNC or
178 * The logic is: Fail if we're at the end-of-string; otherwise if the target is
179 * utf8 and a variant, load the swash if necessary and test using the utf8
180 * test. Advance to the next character if test is ok, otherwise fail; If not
181 * utf8 or an invariant under utf8, use the non-utf8 test, and fail if it
182 * fails, or advance to the next character */
184 #define _CCC_TRY_CODE(POS_OR_NEG, FUNC, UTF8_TEST, CLASS, STR) \
185 if (locinput >= PL_regeol) { \
188 if (utf8_target && UTF8_IS_CONTINUED(nextchr)) { \
189 LOAD_UTF8_CHARCLASS(CLASS, STR); \
190 if (POS_OR_NEG (UTF8_TEST)) { \
193 locinput += PL_utf8skip[nextchr]; \
194 nextchr = UCHARAT(locinput); \
197 if (POS_OR_NEG (FUNC(nextchr))) { \
200 nextchr = UCHARAT(++locinput); \
203 /* Handle the non-locale cases for a character class and its complement. It
204 * calls _CCC_TRY_CODE with a ! to complement the test for the character class.
205 * This is because that code fails when the test succeeds, so we want to have
206 * the test fail so that the code succeeds. The swash is stored in a
207 * predictable PL_ place */
208 #define _CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, \
211 _CCC_TRY_CODE( !, FUNC, \
212 cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), \
213 (U8*)locinput, TRUE)), \
216 _CCC_TRY_CODE( PLACEHOLDER , FUNC, \
217 cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), \
218 (U8*)locinput, TRUE)), \
221 /* Generate the case statements for both locale and non-locale character
222 * classes in regmatch for classes that don't have special unicode semantics.
223 * Locales don't use an immediate swash, but an intermediary special locale
224 * function that is called on the pointer to the current place in the input
225 * string. That function will resolve to needing the same swash. One might
226 * think that because we don't know what the locale will match, we shouldn't
227 * check with the swash loading function that it loaded properly; ie, that we
228 * should use LOAD_UTF8_CHARCLASS_NO_CHECK for those, but what is passed to the
229 * regular LOAD_UTF8_CHARCLASS is in non-locale terms, and so locale is
231 #define CCC_TRY(NAME, NNAME, FUNC, \
232 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
233 NAMEA, NNAMEA, FUNCA, \
236 PL_reg_flags |= RF_tainted; \
237 _CCC_TRY_CODE( !, LCFUNC, LCFUNC_utf8((U8*)locinput), CLASS, STR) \
239 PL_reg_flags |= RF_tainted; \
240 _CCC_TRY_CODE( PLACEHOLDER, LCFUNC, LCFUNC_utf8((U8*)locinput), \
243 if (locinput >= PL_regeol || ! FUNCA(nextchr)) { \
246 /* Matched a utf8-invariant, so don't have to worry about utf8 */ \
247 nextchr = UCHARAT(++locinput); \
250 if (locinput >= PL_regeol || FUNCA(nextchr)) { \
254 locinput += PL_utf8skip[nextchr]; \
255 nextchr = UCHARAT(locinput); \
258 nextchr = UCHARAT(++locinput); \
261 /* Generate the non-locale cases */ \
262 _CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, CLASS, STR)
264 /* This is like CCC_TRY, but has an extra set of parameters for generating case
265 * statements to handle separate Unicode semantics nodes */
266 #define CCC_TRY_U(NAME, NNAME, FUNC, \
267 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
268 NAMEU, NNAMEU, FUNCU, \
269 NAMEA, NNAMEA, FUNCA, \
271 CCC_TRY(NAME, NNAME, FUNC, \
272 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
273 NAMEA, NNAMEA, FUNCA, \
275 _CCC_TRY_NONLOCALE(NAMEU, NNAMEU, FUNCU, CLASS, STR)
277 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
279 /* for use after a quantifier and before an EXACT-like node -- japhy */
280 /* it would be nice to rework regcomp.sym to generate this stuff. sigh
282 * NOTE that *nothing* that affects backtracking should be in here, specifically
283 * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a
284 * node that is in between two EXACT like nodes when ascertaining what the required
285 * "follow" character is. This should probably be moved to regex compile time
286 * although it may be done at run time beause of the REF possibility - more
287 * investigation required. -- demerphq
289 #define JUMPABLE(rn) ( \
291 (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \
293 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
294 OP(rn) == PLUS || OP(rn) == MINMOD || \
296 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
298 #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
300 #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
303 /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
304 we don't need this definition. */
305 #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
306 #define IS_TEXTF(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFU_SS || OP(rn)==EXACTFU_NO_TRIE || OP(rn)==EXACTFA || OP(rn)==EXACTF || OP(rn)==REFF || OP(rn)==NREFF )
307 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
310 /* ... so we use this as its faster. */
311 #define IS_TEXT(rn) ( OP(rn)==EXACT )
312 #define IS_TEXTFU(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFU_SS || OP(rn)==EXACTFU_NO_TRIE || OP(rn) == EXACTFA)
313 #define IS_TEXTF(rn) ( OP(rn)==EXACTF )
314 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
319 Search for mandatory following text node; for lookahead, the text must
320 follow but for lookbehind (rn->flags != 0) we skip to the next step.
322 #define FIND_NEXT_IMPT(rn) STMT_START { \
323 while (JUMPABLE(rn)) { \
324 const OPCODE type = OP(rn); \
325 if (type == SUSPEND || PL_regkind[type] == CURLY) \
326 rn = NEXTOPER(NEXTOPER(rn)); \
327 else if (type == PLUS) \
329 else if (type == IFMATCH) \
330 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
331 else rn += NEXT_OFF(rn); \
336 static void restore_pos(pTHX_ void *arg);
338 #define REGCP_PAREN_ELEMS 4
339 #define REGCP_OTHER_ELEMS 5
340 #define REGCP_FRAME_ELEMS 1
341 /* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
342 * are needed for the regexp context stack bookkeeping. */
345 S_regcppush(pTHX_ I32 parenfloor)
348 const int retval = PL_savestack_ix;
349 const int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS;
350 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
351 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
353 GET_RE_DEBUG_FLAGS_DECL;
355 if (paren_elems_to_push < 0)
356 Perl_croak(aTHX_ "panic: paren_elems_to_push, %i < 0",
357 paren_elems_to_push);
359 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
360 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf
361 " out of range (%lu-%ld)",
362 total_elems, (unsigned long)PL_regsize, (long)parenfloor);
364 SSGROW(total_elems + REGCP_FRAME_ELEMS);
366 for (p = PL_regsize; p > parenfloor; p--) {
367 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
368 SSPUSHINT(PL_regoffs[p].end);
369 SSPUSHINT(PL_regoffs[p].start);
370 SSPUSHPTR(PL_reg_start_tmp[p]);
372 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
373 " saving \\%"UVuf" %"IVdf"(%"IVdf")..%"IVdf"\n",
374 (UV)p, (IV)PL_regoffs[p].start,
375 (IV)(PL_reg_start_tmp[p] - PL_bostr),
376 (IV)PL_regoffs[p].end
379 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
380 SSPUSHPTR(PL_regoffs);
381 SSPUSHINT(PL_regsize);
382 SSPUSHINT(*PL_reglastparen);
383 SSPUSHINT(*PL_reglastcloseparen);
384 SSPUSHPTR(PL_reginput);
385 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
390 /* These are needed since we do not localize EVAL nodes: */
391 #define REGCP_SET(cp) \
393 PerlIO_printf(Perl_debug_log, \
394 " Setting an EVAL scope, savestack=%"IVdf"\n", \
395 (IV)PL_savestack_ix)); \
398 #define REGCP_UNWIND(cp) \
400 if (cp != PL_savestack_ix) \
401 PerlIO_printf(Perl_debug_log, \
402 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
403 (IV)(cp), (IV)PL_savestack_ix)); \
407 S_regcppop(pTHX_ const regexp *rex)
412 GET_RE_DEBUG_FLAGS_DECL;
414 PERL_ARGS_ASSERT_REGCPPOP;
416 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
418 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
419 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
420 input = (char *) SSPOPPTR;
421 *PL_reglastcloseparen = SSPOPINT;
422 *PL_reglastparen = SSPOPINT;
423 PL_regsize = SSPOPINT;
424 PL_regoffs=(regexp_paren_pair *) SSPOPPTR;
426 i -= REGCP_OTHER_ELEMS;
427 /* Now restore the parentheses context. */
428 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
430 U32 paren = (U32)SSPOPINT;
431 PL_reg_start_tmp[paren] = (char *) SSPOPPTR;
432 PL_regoffs[paren].start = SSPOPINT;
434 if (paren <= *PL_reglastparen)
435 PL_regoffs[paren].end = tmps;
437 PerlIO_printf(Perl_debug_log,
438 " restoring \\%"UVuf" to %"IVdf"(%"IVdf")..%"IVdf"%s\n",
439 (UV)paren, (IV)PL_regoffs[paren].start,
440 (IV)(PL_reg_start_tmp[paren] - PL_bostr),
441 (IV)PL_regoffs[paren].end,
442 (paren > *PL_reglastparen ? "(no)" : ""));
446 if (*PL_reglastparen + 1 <= rex->nparens) {
447 PerlIO_printf(Perl_debug_log,
448 " restoring \\%"IVdf"..\\%"IVdf" to undef\n",
449 (IV)(*PL_reglastparen + 1), (IV)rex->nparens);
453 /* It would seem that the similar code in regtry()
454 * already takes care of this, and in fact it is in
455 * a better location to since this code can #if 0-ed out
456 * but the code in regtry() is needed or otherwise tests
457 * requiring null fields (pat.t#187 and split.t#{13,14}
458 * (as of patchlevel 7877) will fail. Then again,
459 * this code seems to be necessary or otherwise
460 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
461 * --jhi updated by dapm */
462 for (i = *PL_reglastparen + 1; i <= rex->nparens; i++) {
464 PL_regoffs[i].start = -1;
465 PL_regoffs[i].end = -1;
471 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
474 * pregexec and friends
477 #ifndef PERL_IN_XSUB_RE
479 - pregexec - match a regexp against a string
482 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, register char *strend,
483 char *strbeg, I32 minend, SV *screamer, U32 nosave)
484 /* strend: pointer to null at end of string */
485 /* strbeg: real beginning of string */
486 /* minend: end of match must be >=minend after stringarg. */
487 /* nosave: For optimizations. */
489 PERL_ARGS_ASSERT_PREGEXEC;
492 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
493 nosave ? 0 : REXEC_COPY_STR);
498 * Need to implement the following flags for reg_anch:
500 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
502 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
503 * INTUIT_AUTORITATIVE_ML
504 * INTUIT_ONCE_NOML - Intuit can match in one location only.
507 * Another flag for this function: SECOND_TIME (so that float substrs
508 * with giant delta may be not rechecked).
511 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
513 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
514 Otherwise, only SvCUR(sv) is used to get strbeg. */
516 /* XXXX We assume that strpos is strbeg unless sv. */
518 /* XXXX Some places assume that there is a fixed substring.
519 An update may be needed if optimizer marks as "INTUITable"
520 RExen without fixed substrings. Similarly, it is assumed that
521 lengths of all the strings are no more than minlen, thus they
522 cannot come from lookahead.
523 (Or minlen should take into account lookahead.)
524 NOTE: Some of this comment is not correct. minlen does now take account
525 of lookahead/behind. Further research is required. -- demerphq
529 /* A failure to find a constant substring means that there is no need to make
530 an expensive call to REx engine, thus we celebrate a failure. Similarly,
531 finding a substring too deep into the string means that less calls to
532 regtry() should be needed.
534 REx compiler's optimizer found 4 possible hints:
535 a) Anchored substring;
537 c) Whether we are anchored (beginning-of-line or \G);
538 d) First node (of those at offset 0) which may distinguish positions;
539 We use a)b)d) and multiline-part of c), and try to find a position in the
540 string which does not contradict any of them.
543 /* Most of decisions we do here should have been done at compile time.
544 The nodes of the REx which we used for the search should have been
545 deleted from the finite automaton. */
548 Perl_re_intuit_start(pTHX_ REGEXP * const rx, SV *sv, char *strpos,
549 char *strend, const U32 flags, re_scream_pos_data *data)
552 struct regexp *const prog = (struct regexp *)SvANY(rx);
553 register I32 start_shift = 0;
554 /* Should be nonnegative! */
555 register I32 end_shift = 0;
560 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
562 register char *other_last = NULL; /* other substr checked before this */
563 char *check_at = NULL; /* check substr found at this pos */
564 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
565 RXi_GET_DECL(prog,progi);
567 const char * const i_strpos = strpos;
569 GET_RE_DEBUG_FLAGS_DECL;
571 PERL_ARGS_ASSERT_RE_INTUIT_START;
573 RX_MATCH_UTF8_set(rx,utf8_target);
576 PL_reg_flags |= RF_utf8;
579 debug_start_match(rx, utf8_target, strpos, strend,
580 sv ? "Guessing start of match in sv for"
581 : "Guessing start of match in string for");
584 /* CHR_DIST() would be more correct here but it makes things slow. */
585 if (prog->minlen > strend - strpos) {
586 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
587 "String too short... [re_intuit_start]\n"));
591 strbeg = (sv && SvPOK(sv)) ? strend - SvCUR(sv) : strpos;
594 if (!prog->check_utf8 && prog->check_substr)
595 to_utf8_substr(prog);
596 check = prog->check_utf8;
598 if (!prog->check_substr && prog->check_utf8)
599 to_byte_substr(prog);
600 check = prog->check_substr;
602 if (check == &PL_sv_undef) {
603 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
604 "Non-utf8 string cannot match utf8 check string\n"));
607 if (prog->extflags & RXf_ANCH) { /* Match at beg-of-str or after \n */
608 ml_anch = !( (prog->extflags & RXf_ANCH_SINGLE)
609 || ( (prog->extflags & RXf_ANCH_BOL)
610 && !multiline ) ); /* Check after \n? */
613 if ( !(prog->extflags & RXf_ANCH_GPOS) /* Checked by the caller */
614 && !(prog->intflags & PREGf_IMPLICIT) /* not a real BOL */
615 /* SvCUR is not set on references: SvRV and SvPVX_const overlap */
617 && (strpos != strbeg)) {
618 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
621 if (prog->check_offset_min == prog->check_offset_max &&
622 !(prog->extflags & RXf_CANY_SEEN)) {
623 /* Substring at constant offset from beg-of-str... */
626 s = HOP3c(strpos, prog->check_offset_min, strend);
629 slen = SvCUR(check); /* >= 1 */
631 if ( strend - s > slen || strend - s < slen - 1
632 || (strend - s == slen && strend[-1] != '\n')) {
633 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
636 /* Now should match s[0..slen-2] */
638 if (slen && (*SvPVX_const(check) != *s
640 && memNE(SvPVX_const(check), s, slen)))) {
642 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
646 else if (*SvPVX_const(check) != *s
647 || ((slen = SvCUR(check)) > 1
648 && memNE(SvPVX_const(check), s, slen)))
651 goto success_at_start;
654 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
656 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
657 end_shift = prog->check_end_shift;
660 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
661 - (SvTAIL(check) != 0);
662 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
664 if (end_shift < eshift)
668 else { /* Can match at random position */
671 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
672 end_shift = prog->check_end_shift;
674 /* end shift should be non negative here */
677 #ifdef QDEBUGGING /* 7/99: reports of failure (with the older version) */
679 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
680 (IV)end_shift, RX_PRECOMP(prog));
684 /* Find a possible match in the region s..strend by looking for
685 the "check" substring in the region corrected by start/end_shift. */
688 I32 srch_start_shift = start_shift;
689 I32 srch_end_shift = end_shift;
690 if (srch_start_shift < 0 && strbeg - s > srch_start_shift) {
691 srch_end_shift -= ((strbeg - s) - srch_start_shift);
692 srch_start_shift = strbeg - s;
694 DEBUG_OPTIMISE_MORE_r({
695 PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
696 (IV)prog->check_offset_min,
697 (IV)srch_start_shift,
699 (IV)prog->check_end_shift);
702 if ((flags & REXEC_SCREAM) && SvSCREAM(sv)) {
703 I32 p = -1; /* Internal iterator of scream. */
704 I32 * const pp = data ? data->scream_pos : &p;
708 assert(SvMAGICAL(sv));
709 mg = mg_find(sv, PERL_MAGIC_study);
712 if (mg->mg_private == 1) {
713 found = ((U8 *)mg->mg_ptr)[BmRARE(check)] != (U8)~0;
714 } else if (mg->mg_private == 2) {
715 found = ((U16 *)mg->mg_ptr)[BmRARE(check)] != (U16)~0;
717 assert (mg->mg_private == 4);
718 found = ((U32 *)mg->mg_ptr)[BmRARE(check)] != (U32)~0;
722 || ( BmRARE(check) == '\n'
723 && (BmPREVIOUS(check) == SvCUR(check) - 1)
725 s = screaminstr(sv, check,
726 srch_start_shift + (s - strbeg), srch_end_shift, pp, 0);
729 /* we may be pointing at the wrong string */
730 if (s && RXp_MATCH_COPIED(prog))
731 s = strbeg + (s - SvPVX_const(sv));
733 *data->scream_olds = s;
738 if (prog->extflags & RXf_CANY_SEEN) {
739 start_point= (U8*)(s + srch_start_shift);
740 end_point= (U8*)(strend - srch_end_shift);
742 start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
743 end_point= HOP3(strend, -srch_end_shift, strbeg);
745 DEBUG_OPTIMISE_MORE_r({
746 PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
747 (int)(end_point - start_point),
748 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
752 s = fbm_instr( start_point, end_point,
753 check, multiline ? FBMrf_MULTILINE : 0);
756 /* Update the count-of-usability, remove useless subpatterns,
760 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
761 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
762 PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s",
763 (s ? "Found" : "Did not find"),
764 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
765 ? "anchored" : "floating"),
768 (s ? " at offset " : "...\n") );
773 /* Finish the diagnostic message */
774 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
776 /* XXX dmq: first branch is for positive lookbehind...
777 Our check string is offset from the beginning of the pattern.
778 So we need to do any stclass tests offset forward from that
787 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
788 Start with the other substr.
789 XXXX no SCREAM optimization yet - and a very coarse implementation
790 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
791 *always* match. Probably should be marked during compile...
792 Probably it is right to do no SCREAM here...
795 if (utf8_target ? (prog->float_utf8 && prog->anchored_utf8)
796 : (prog->float_substr && prog->anchored_substr))
798 /* Take into account the "other" substring. */
799 /* XXXX May be hopelessly wrong for UTF... */
802 if (check == (utf8_target ? prog->float_utf8 : prog->float_substr)) {
805 char * const last = HOP3c(s, -start_shift, strbeg);
807 char * const saved_s = s;
810 t = s - prog->check_offset_max;
811 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
813 || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos))
818 t = HOP3c(t, prog->anchored_offset, strend);
819 if (t < other_last) /* These positions already checked */
821 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
824 /* XXXX It is not documented what units *_offsets are in.
825 We assume bytes, but this is clearly wrong.
826 Meaning this code needs to be carefully reviewed for errors.
830 /* On end-of-str: see comment below. */
831 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
832 if (must == &PL_sv_undef) {
834 DEBUG_r(must = prog->anchored_utf8); /* for debug */
839 HOP3(HOP3(last1, prog->anchored_offset, strend)
840 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
842 multiline ? FBMrf_MULTILINE : 0
845 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
846 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
847 PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s",
848 (s ? "Found" : "Contradicts"),
849 quoted, RE_SV_TAIL(must));
854 if (last1 >= last2) {
855 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
856 ", giving up...\n"));
859 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
860 ", trying floating at offset %ld...\n",
861 (long)(HOP3c(saved_s, 1, strend) - i_strpos)));
862 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
863 s = HOP3c(last, 1, strend);
867 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
868 (long)(s - i_strpos)));
869 t = HOP3c(s, -prog->anchored_offset, strbeg);
870 other_last = HOP3c(s, 1, strend);
878 else { /* Take into account the floating substring. */
880 char * const saved_s = s;
883 t = HOP3c(s, -start_shift, strbeg);
885 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
886 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
887 last = HOP3c(t, prog->float_max_offset, strend);
888 s = HOP3c(t, prog->float_min_offset, strend);
891 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
892 must = utf8_target ? prog->float_utf8 : prog->float_substr;
893 /* fbm_instr() takes into account exact value of end-of-str
894 if the check is SvTAIL(ed). Since false positives are OK,
895 and end-of-str is not later than strend we are OK. */
896 if (must == &PL_sv_undef) {
898 DEBUG_r(must = prog->float_utf8); /* for debug message */
901 s = fbm_instr((unsigned char*)s,
902 (unsigned char*)last + SvCUR(must)
904 must, multiline ? FBMrf_MULTILINE : 0);
906 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
907 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
908 PerlIO_printf(Perl_debug_log, "%s floating substr %s%s",
909 (s ? "Found" : "Contradicts"),
910 quoted, RE_SV_TAIL(must));
914 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
915 ", giving up...\n"));
918 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
919 ", trying anchored starting at offset %ld...\n",
920 (long)(saved_s + 1 - i_strpos)));
922 s = HOP3c(t, 1, strend);
926 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
927 (long)(s - i_strpos)));
928 other_last = s; /* Fix this later. --Hugo */
938 t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
940 DEBUG_OPTIMISE_MORE_r(
941 PerlIO_printf(Perl_debug_log,
942 "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
943 (IV)prog->check_offset_min,
944 (IV)prog->check_offset_max,
952 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
954 || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos)))
957 /* Fixed substring is found far enough so that the match
958 cannot start at strpos. */
960 if (ml_anch && t[-1] != '\n') {
961 /* Eventually fbm_*() should handle this, but often
962 anchored_offset is not 0, so this check will not be wasted. */
963 /* XXXX In the code below we prefer to look for "^" even in
964 presence of anchored substrings. And we search even
965 beyond the found float position. These pessimizations
966 are historical artefacts only. */
968 while (t < strend - prog->minlen) {
970 if (t < check_at - prog->check_offset_min) {
971 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
972 /* Since we moved from the found position,
973 we definitely contradict the found anchored
974 substr. Due to the above check we do not
975 contradict "check" substr.
976 Thus we can arrive here only if check substr
977 is float. Redo checking for "other"=="fixed".
980 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
981 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
982 goto do_other_anchored;
984 /* We don't contradict the found floating substring. */
985 /* XXXX Why not check for STCLASS? */
987 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
988 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
991 /* Position contradicts check-string */
992 /* XXXX probably better to look for check-string
993 than for "\n", so one should lower the limit for t? */
994 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
995 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
996 other_last = strpos = s = t + 1;
1001 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
1002 PL_colors[0], PL_colors[1]));
1006 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
1007 PL_colors[0], PL_colors[1]));
1011 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
1014 /* The found string does not prohibit matching at strpos,
1015 - no optimization of calling REx engine can be performed,
1016 unless it was an MBOL and we are not after MBOL,
1017 or a future STCLASS check will fail this. */
1019 /* Even in this situation we may use MBOL flag if strpos is offset
1020 wrt the start of the string. */
1021 if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
1022 && (strpos != strbeg) && strpos[-1] != '\n'
1023 /* May be due to an implicit anchor of m{.*foo} */
1024 && !(prog->intflags & PREGf_IMPLICIT))
1029 DEBUG_EXECUTE_r( if (ml_anch)
1030 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
1031 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
1034 if (!(prog->intflags & PREGf_NAUGHTY) /* XXXX If strpos moved? */
1036 prog->check_utf8 /* Could be deleted already */
1037 && --BmUSEFUL(prog->check_utf8) < 0
1038 && (prog->check_utf8 == prog->float_utf8)
1040 prog->check_substr /* Could be deleted already */
1041 && --BmUSEFUL(prog->check_substr) < 0
1042 && (prog->check_substr == prog->float_substr)
1045 /* If flags & SOMETHING - do not do it many times on the same match */
1046 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
1047 /* XXX Does the destruction order has to change with utf8_target? */
1048 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1049 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1050 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1051 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1052 check = NULL; /* abort */
1054 /* XXXX If the check string was an implicit check MBOL, then we need to unset the relevant flag
1055 see http://bugs.activestate.com/show_bug.cgi?id=87173 */
1056 if (prog->intflags & PREGf_IMPLICIT)
1057 prog->extflags &= ~RXf_ANCH_MBOL;
1058 /* XXXX This is a remnant of the old implementation. It
1059 looks wasteful, since now INTUIT can use many
1060 other heuristics. */
1061 prog->extflags &= ~RXf_USE_INTUIT;
1062 /* XXXX What other flags might need to be cleared in this branch? */
1068 /* Last resort... */
1069 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
1070 /* trie stclasses are too expensive to use here, we are better off to
1071 leave it to regmatch itself */
1072 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
1073 /* minlen == 0 is possible if regstclass is \b or \B,
1074 and the fixed substr is ''$.
1075 Since minlen is already taken into account, s+1 is before strend;
1076 accidentally, minlen >= 1 guaranties no false positives at s + 1
1077 even for \b or \B. But (minlen? 1 : 0) below assumes that
1078 regstclass does not come from lookahead... */
1079 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
1080 This leaves EXACTF-ish only, which are dealt with in find_byclass(). */
1081 const U8* const str = (U8*)STRING(progi->regstclass);
1082 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
1083 ? CHR_DIST(str+STR_LEN(progi->regstclass), str)
1086 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
1087 endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend);
1088 else if (prog->float_substr || prog->float_utf8)
1089 endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend);
1093 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" s: %"IVdf" endpos: %"IVdf"\n",
1094 (IV)start_shift, (IV)(check_at - strbeg), (IV)(s - strbeg), (IV)(endpos - strbeg)));
1097 s = find_byclass(prog, progi->regstclass, s, endpos, NULL);
1100 const char *what = NULL;
1102 if (endpos == strend) {
1103 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1104 "Could not match STCLASS...\n") );
1107 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1108 "This position contradicts STCLASS...\n") );
1109 if ((prog->extflags & RXf_ANCH) && !ml_anch)
1111 /* Contradict one of substrings */
1112 if (prog->anchored_substr || prog->anchored_utf8) {
1113 if ((utf8_target ? prog->anchored_utf8 : prog->anchored_substr) == check) {
1114 DEBUG_EXECUTE_r( what = "anchored" );
1116 s = HOP3c(t, 1, strend);
1117 if (s + start_shift + end_shift > strend) {
1118 /* XXXX Should be taken into account earlier? */
1119 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1120 "Could not match STCLASS...\n") );
1125 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1126 "Looking for %s substr starting at offset %ld...\n",
1127 what, (long)(s + start_shift - i_strpos)) );
1130 /* Have both, check_string is floating */
1131 if (t + start_shift >= check_at) /* Contradicts floating=check */
1132 goto retry_floating_check;
1133 /* Recheck anchored substring, but not floating... */
1137 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1138 "Looking for anchored substr starting at offset %ld...\n",
1139 (long)(other_last - i_strpos)) );
1140 goto do_other_anchored;
1142 /* Another way we could have checked stclass at the
1143 current position only: */
1148 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1149 "Looking for /%s^%s/m starting at offset %ld...\n",
1150 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
1153 if (!(utf8_target ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
1155 /* Check is floating substring. */
1156 retry_floating_check:
1157 t = check_at - start_shift;
1158 DEBUG_EXECUTE_r( what = "floating" );
1159 goto hop_and_restart;
1162 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1163 "By STCLASS: moving %ld --> %ld\n",
1164 (long)(t - i_strpos), (long)(s - i_strpos))
1168 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1169 "Does not contradict STCLASS...\n");
1174 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
1175 PL_colors[4], (check ? "Guessed" : "Giving up"),
1176 PL_colors[5], (long)(s - i_strpos)) );
1179 fail_finish: /* Substring not found */
1180 if (prog->check_substr || prog->check_utf8) /* could be removed already */
1181 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
1183 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
1184 PL_colors[4], PL_colors[5]));
1188 #define DECL_TRIE_TYPE(scan) \
1189 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold } \
1190 trie_type = (scan->flags != EXACT) \
1191 ? (utf8_target ? trie_utf8_fold : (UTF_PATTERN ? trie_latin_utf8_fold : trie_plain)) \
1192 : (utf8_target ? trie_utf8 : trie_plain)
1194 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, \
1195 uvc, charid, foldlen, foldbuf, uniflags) STMT_START { \
1196 switch (trie_type) { \
1197 case trie_utf8_fold: \
1198 if ( foldlen>0 ) { \
1199 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
1204 uvc = to_utf8_fold( (U8 *) uc, foldbuf, &foldlen ); \
1205 len = UTF8SKIP(uc); \
1206 foldlen -= UNISKIP( uvc ); \
1207 uscan = foldbuf + UNISKIP( uvc ); \
1210 case trie_latin_utf8_fold: \
1211 if ( foldlen>0 ) { \
1212 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
1218 uvc = to_uni_fold( *(U8*)uc, foldbuf, &foldlen ); \
1219 foldlen -= UNISKIP( uvc ); \
1220 uscan = foldbuf + UNISKIP( uvc ); \
1224 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
1231 charid = trie->charmap[ uvc ]; \
1235 if (widecharmap) { \
1236 SV** const svpp = hv_fetch(widecharmap, \
1237 (char*)&uvc, sizeof(UV), 0); \
1239 charid = (U16)SvIV(*svpp); \
1244 #define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1248 && (ln == 1 || folder(s, pat_string, ln)) \
1249 && (!reginfo || regtry(reginfo, &s)) ) \
1255 #define REXEC_FBC_UTF8_SCAN(CoDe) \
1257 while (s + (uskip = UTF8SKIP(s)) <= strend) { \
1263 #define REXEC_FBC_SCAN(CoDe) \
1265 while (s < strend) { \
1271 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1272 REXEC_FBC_UTF8_SCAN( \
1274 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1283 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1286 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1295 #define REXEC_FBC_TRYIT \
1296 if ((!reginfo || regtry(reginfo, &s))) \
1299 #define REXEC_FBC_CSCAN(CoNdUtF8,CoNd) \
1300 if (utf8_target) { \
1301 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1304 REXEC_FBC_CLASS_SCAN(CoNd); \
1307 #define REXEC_FBC_CSCAN_PRELOAD(UtFpReLoAd,CoNdUtF8,CoNd) \
1308 if (utf8_target) { \
1310 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1313 REXEC_FBC_CLASS_SCAN(CoNd); \
1316 #define REXEC_FBC_CSCAN_TAINT(CoNdUtF8,CoNd) \
1317 PL_reg_flags |= RF_tainted; \
1318 if (utf8_target) { \
1319 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1322 REXEC_FBC_CLASS_SCAN(CoNd); \
1325 #define DUMP_EXEC_POS(li,s,doutf8) \
1326 dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8)
1329 #define UTF8_NOLOAD(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1330 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1331 tmp = TEST_NON_UTF8(tmp); \
1332 REXEC_FBC_UTF8_SCAN( \
1333 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1342 #define UTF8_LOAD(TeSt1_UtF8, TeSt2_UtF8, IF_SUCCESS, IF_FAIL) \
1343 if (s == PL_bostr) { \
1347 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr); \
1348 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT); \
1351 LOAD_UTF8_CHARCLASS_ALNUM(); \
1352 REXEC_FBC_UTF8_SCAN( \
1353 if (tmp == ! (TeSt2_UtF8)) { \
1362 /* The only difference between the BOUND and NBOUND cases is that
1363 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1364 * NBOUND. This is accomplished by passing it in either the if or else clause,
1365 * with the other one being empty */
1366 #define FBC_BOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1367 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1369 #define FBC_BOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1370 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1372 #define FBC_NBOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1373 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1375 #define FBC_NBOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1376 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1379 /* Common to the BOUND and NBOUND cases. Unfortunately the UTF8 tests need to
1380 * be passed in completely with the variable name being tested, which isn't
1381 * such a clean interface, but this is easier to read than it was before. We
1382 * are looking for the boundary (or non-boundary between a word and non-word
1383 * character. The utf8 and non-utf8 cases have the same logic, but the details
1384 * must be different. Find the "wordness" of the character just prior to this
1385 * one, and compare it with the wordness of this one. If they differ, we have
1386 * a boundary. At the beginning of the string, pretend that the previous
1387 * character was a new-line */
1388 #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1389 if (utf8_target) { \
1392 else { /* Not utf8 */ \
1393 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1394 tmp = TEST_NON_UTF8(tmp); \
1396 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1405 if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, &s))) \
1408 /* We know what class REx starts with. Try to find this position... */
1409 /* if reginfo is NULL, its a dryrun */
1410 /* annoyingly all the vars in this routine have different names from their counterparts
1411 in regmatch. /grrr */
1414 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1415 const char *strend, regmatch_info *reginfo)
1418 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1419 char *pat_string; /* The pattern's exactish string */
1420 char *pat_end; /* ptr to end char of pat_string */
1421 re_fold_t folder; /* Function for computing non-utf8 folds */
1422 const U8 *fold_array; /* array for folding ords < 256 */
1425 register STRLEN uskip;
1429 register I32 tmp = 1; /* Scratch variable? */
1430 register const bool utf8_target = PL_reg_match_utf8;
1431 UV utf8_fold_flags = 0;
1432 RXi_GET_DECL(prog,progi);
1434 PERL_ARGS_ASSERT_FIND_BYCLASS;
1436 /* We know what class it must start with. */
1440 if (utf8_target || OP(c) == ANYOFV) {
1441 STRLEN inclasslen = strend - s;
1442 REXEC_FBC_UTF8_CLASS_SCAN(
1443 reginclass(prog, c, (U8*)s, &inclasslen, utf8_target));
1446 REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s));
1451 if (tmp && (!reginfo || regtry(reginfo, &s)))
1459 if (UTF_PATTERN || utf8_target) {
1460 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1461 goto do_exactf_utf8;
1463 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1464 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1465 goto do_exactf_non_utf8; /* isn't dealt with by these */
1470 /* regcomp.c already folded this if pattern is in UTF-8 */
1471 utf8_fold_flags = 0;
1472 goto do_exactf_utf8;
1474 fold_array = PL_fold;
1476 goto do_exactf_non_utf8;
1479 if (UTF_PATTERN || utf8_target) {
1480 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
1481 goto do_exactf_utf8;
1483 fold_array = PL_fold_locale;
1484 folder = foldEQ_locale;
1485 goto do_exactf_non_utf8;
1489 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
1491 goto do_exactf_utf8;
1493 case EXACTFU_NO_TRIE:
1495 if (UTF_PATTERN || utf8_target) {
1496 utf8_fold_flags = (UTF_PATTERN) ? FOLDEQ_S2_ALREADY_FOLDED : 0;
1497 goto do_exactf_utf8;
1500 /* Any 'ss' in the pattern should have been replaced by regcomp,
1501 * so we don't have to worry here about this single special case
1502 * in the Latin1 range */
1503 fold_array = PL_fold_latin1;
1504 folder = foldEQ_latin1;
1508 do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there
1509 are no glitches with fold-length differences
1510 between the target string and pattern */
1512 /* The idea in the non-utf8 EXACTF* cases is to first find the
1513 * first character of the EXACTF* node and then, if necessary,
1514 * case-insensitively compare the full text of the node. c1 is the
1515 * first character. c2 is its fold. This logic will not work for
1516 * Unicode semantics and the german sharp ss, which hence should
1517 * not be compiled into a node that gets here. */
1518 pat_string = STRING(c);
1519 ln = STR_LEN(c); /* length to match in octets/bytes */
1521 /* We know that we have to match at least 'ln' bytes (which is the
1522 * same as characters, since not utf8). If we have to match 3
1523 * characters, and there are only 2 availabe, we know without
1524 * trying that it will fail; so don't start a match past the
1525 * required minimum number from the far end */
1526 e = HOP3c(strend, -((I32)ln), s);
1528 if (!reginfo && e < s) {
1529 e = s; /* Due to minlen logic of intuit() */
1533 c2 = fold_array[c1];
1534 if (c1 == c2) { /* If char and fold are the same */
1535 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1538 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1547 /* If one of the operands is in utf8, we can't use the simpler
1548 * folding above, due to the fact that many different characters
1549 * can have the same fold, or portion of a fold, or different-
1551 pat_string = STRING(c);
1552 ln = STR_LEN(c); /* length to match in octets/bytes */
1553 pat_end = pat_string + ln;
1554 lnc = (UTF_PATTERN) /* length to match in characters */
1555 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
1558 /* We have 'lnc' characters to match in the pattern, but because of
1559 * multi-character folding, each character in the target can match
1560 * up to 3 characters (Unicode guarantees it will never exceed
1561 * this) if it is utf8-encoded; and up to 2 if not (based on the
1562 * fact that the Latin 1 folds are already determined, and the
1563 * only multi-char fold in that range is the sharp-s folding to
1564 * 'ss'. Thus, a pattern character can match as little as 1/3 of a
1565 * string character. Adjust lnc accordingly, rounding up, so that
1566 * if we need to match at least 4+1/3 chars, that really is 5. */
1567 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
1568 lnc = (lnc + expansion - 1) / expansion;
1570 /* As in the non-UTF8 case, if we have to match 3 characters, and
1571 * only 2 are left, it's guaranteed to fail, so don't start a
1572 * match that would require us to go beyond the end of the string
1574 e = HOP3c(strend, -((I32)lnc), s);
1576 if (!reginfo && e < s) {
1577 e = s; /* Due to minlen logic of intuit() */
1580 /* XXX Note that we could recalculate e to stop the loop earlier,
1581 * as the worst case expansion above will rarely be met, and as we
1582 * go along we would usually find that e moves further to the left.
1583 * This would happen only after we reached the point in the loop
1584 * where if there were no expansion we should fail. Unclear if
1585 * worth the expense */
1588 char *my_strend= (char *)strend;
1589 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
1590 pat_string, NULL, ln, cBOOL(UTF_PATTERN), utf8_fold_flags)
1591 && (!reginfo || regtry(reginfo, &s)) )
1595 s += (utf8_target) ? UTF8SKIP(s) : 1;
1600 PL_reg_flags |= RF_tainted;
1601 FBC_BOUND(isALNUM_LC,
1602 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1603 isALNUM_LC_utf8((U8*)s));
1606 PL_reg_flags |= RF_tainted;
1607 FBC_NBOUND(isALNUM_LC,
1608 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1609 isALNUM_LC_utf8((U8*)s));
1612 FBC_BOUND(isWORDCHAR,
1614 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1617 FBC_BOUND_NOLOAD(isWORDCHAR_A,
1619 isWORDCHAR_A((U8*)s));
1622 FBC_NBOUND(isWORDCHAR,
1624 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1627 FBC_NBOUND_NOLOAD(isWORDCHAR_A,
1629 isWORDCHAR_A((U8*)s));
1632 FBC_BOUND(isWORDCHAR_L1,
1634 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1637 FBC_NBOUND(isWORDCHAR_L1,
1639 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1642 REXEC_FBC_CSCAN_TAINT(
1643 isALNUM_LC_utf8((U8*)s),
1648 REXEC_FBC_CSCAN_PRELOAD(
1649 LOAD_UTF8_CHARCLASS_ALNUM(),
1650 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1651 isWORDCHAR_L1((U8) *s)
1655 REXEC_FBC_CSCAN_PRELOAD(
1656 LOAD_UTF8_CHARCLASS_ALNUM(),
1657 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1662 /* Don't need to worry about utf8, as it can match only a single
1663 * byte invariant character */
1664 REXEC_FBC_CLASS_SCAN( isWORDCHAR_A(*s));
1667 REXEC_FBC_CSCAN_PRELOAD(
1668 LOAD_UTF8_CHARCLASS_ALNUM(),
1669 !swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1670 ! isWORDCHAR_L1((U8) *s)
1674 REXEC_FBC_CSCAN_PRELOAD(
1675 LOAD_UTF8_CHARCLASS_ALNUM(),
1676 !swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target),
1687 REXEC_FBC_CSCAN_TAINT(
1688 !isALNUM_LC_utf8((U8*)s),
1693 REXEC_FBC_CSCAN_PRELOAD(
1694 LOAD_UTF8_CHARCLASS_SPACE(),
1695 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target),
1700 REXEC_FBC_CSCAN_PRELOAD(
1701 LOAD_UTF8_CHARCLASS_SPACE(),
1702 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target),
1707 /* Don't need to worry about utf8, as it can match only a single
1708 * byte invariant character */
1709 REXEC_FBC_CLASS_SCAN( isSPACE_A(*s));
1712 REXEC_FBC_CSCAN_TAINT(
1713 isSPACE_LC_utf8((U8*)s),
1718 REXEC_FBC_CSCAN_PRELOAD(
1719 LOAD_UTF8_CHARCLASS_SPACE(),
1720 !( *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)),
1721 ! isSPACE_L1((U8) *s)
1725 REXEC_FBC_CSCAN_PRELOAD(
1726 LOAD_UTF8_CHARCLASS_SPACE(),
1727 !(*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)),
1738 REXEC_FBC_CSCAN_TAINT(
1739 !isSPACE_LC_utf8((U8*)s),
1744 REXEC_FBC_CSCAN_PRELOAD(
1745 LOAD_UTF8_CHARCLASS_DIGIT(),
1746 swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1751 /* Don't need to worry about utf8, as it can match only a single
1752 * byte invariant character */
1753 REXEC_FBC_CLASS_SCAN( isDIGIT_A(*s));
1756 REXEC_FBC_CSCAN_TAINT(
1757 isDIGIT_LC_utf8((U8*)s),
1762 REXEC_FBC_CSCAN_PRELOAD(
1763 LOAD_UTF8_CHARCLASS_DIGIT(),
1764 !swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1775 REXEC_FBC_CSCAN_TAINT(
1776 !isDIGIT_LC_utf8((U8*)s),
1783 is_LNBREAK_latin1(s)
1795 !is_VERTWS_latin1(s)
1801 is_HORIZWS_latin1(s)
1806 !is_HORIZWS_utf8(s),
1807 !is_HORIZWS_latin1(s)
1814 /* what trie are we using right now */
1816 = (reg_ac_data*)progi->data->data[ ARG( c ) ];
1818 = (reg_trie_data*)progi->data->data[ aho->trie ];
1819 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
1821 const char *last_start = strend - trie->minlen;
1823 const char *real_start = s;
1825 STRLEN maxlen = trie->maxlen;
1827 U8 **points; /* map of where we were in the input string
1828 when reading a given char. For ASCII this
1829 is unnecessary overhead as the relationship
1830 is always 1:1, but for Unicode, especially
1831 case folded Unicode this is not true. */
1832 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1836 GET_RE_DEBUG_FLAGS_DECL;
1838 /* We can't just allocate points here. We need to wrap it in
1839 * an SV so it gets freed properly if there is a croak while
1840 * running the match */
1843 sv_points=newSV(maxlen * sizeof(U8 *));
1844 SvCUR_set(sv_points,
1845 maxlen * sizeof(U8 *));
1846 SvPOK_on(sv_points);
1847 sv_2mortal(sv_points);
1848 points=(U8**)SvPV_nolen(sv_points );
1849 if ( trie_type != trie_utf8_fold
1850 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1853 bitmap=(U8*)trie->bitmap;
1855 bitmap=(U8*)ANYOF_BITMAP(c);
1857 /* this is the Aho-Corasick algorithm modified a touch
1858 to include special handling for long "unknown char"
1859 sequences. The basic idea being that we use AC as long
1860 as we are dealing with a possible matching char, when
1861 we encounter an unknown char (and we have not encountered
1862 an accepting state) we scan forward until we find a legal
1864 AC matching is basically that of trie matching, except
1865 that when we encounter a failing transition, we fall back
1866 to the current states "fail state", and try the current char
1867 again, a process we repeat until we reach the root state,
1868 state 1, or a legal transition. If we fail on the root state
1869 then we can either terminate if we have reached an accepting
1870 state previously, or restart the entire process from the beginning
1874 while (s <= last_start) {
1875 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1883 U8 *uscan = (U8*)NULL;
1884 U8 *leftmost = NULL;
1886 U32 accepted_word= 0;
1890 while ( state && uc <= (U8*)strend ) {
1892 U32 word = aho->states[ state ].wordnum;
1896 DEBUG_TRIE_EXECUTE_r(
1897 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1898 dump_exec_pos( (char *)uc, c, strend, real_start,
1899 (char *)uc, utf8_target );
1900 PerlIO_printf( Perl_debug_log,
1901 " Scanning for legal start char...\n");
1905 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1909 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1915 if (uc >(U8*)last_start) break;
1919 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
1920 if (!leftmost || lpos < leftmost) {
1921 DEBUG_r(accepted_word=word);
1927 points[pointpos++ % maxlen]= uc;
1928 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
1929 uscan, len, uvc, charid, foldlen,
1931 DEBUG_TRIE_EXECUTE_r({
1932 dump_exec_pos( (char *)uc, c, strend, real_start,
1934 PerlIO_printf(Perl_debug_log,
1935 " Charid:%3u CP:%4"UVxf" ",
1941 word = aho->states[ state ].wordnum;
1943 base = aho->states[ state ].trans.base;
1945 DEBUG_TRIE_EXECUTE_r({
1947 dump_exec_pos( (char *)uc, c, strend, real_start,
1949 PerlIO_printf( Perl_debug_log,
1950 "%sState: %4"UVxf", word=%"UVxf,
1951 failed ? " Fail transition to " : "",
1952 (UV)state, (UV)word);
1958 ( ((offset = base + charid
1959 - 1 - trie->uniquecharcount)) >= 0)
1960 && ((U32)offset < trie->lasttrans)
1961 && trie->trans[offset].check == state
1962 && (tmp=trie->trans[offset].next))
1964 DEBUG_TRIE_EXECUTE_r(
1965 PerlIO_printf( Perl_debug_log," - legal\n"));
1970 DEBUG_TRIE_EXECUTE_r(
1971 PerlIO_printf( Perl_debug_log," - fail\n"));
1973 state = aho->fail[state];
1977 /* we must be accepting here */
1978 DEBUG_TRIE_EXECUTE_r(
1979 PerlIO_printf( Perl_debug_log," - accepting\n"));
1988 if (!state) state = 1;
1991 if ( aho->states[ state ].wordnum ) {
1992 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
1993 if (!leftmost || lpos < leftmost) {
1994 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
1999 s = (char*)leftmost;
2000 DEBUG_TRIE_EXECUTE_r({
2002 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
2003 (UV)accepted_word, (IV)(s - real_start)
2006 if (!reginfo || regtry(reginfo, &s)) {
2012 DEBUG_TRIE_EXECUTE_r({
2013 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
2016 DEBUG_TRIE_EXECUTE_r(
2017 PerlIO_printf( Perl_debug_log,"No match.\n"));
2026 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
2036 - regexec_flags - match a regexp against a string
2039 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, register char *strend,
2040 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
2041 /* strend: pointer to null at end of string */
2042 /* strbeg: real beginning of string */
2043 /* minend: end of match must be >=minend after stringarg. */
2044 /* data: May be used for some additional optimizations.
2045 Currently its only used, with a U32 cast, for transmitting
2046 the ganch offset when doing a /g match. This will change */
2047 /* nosave: For optimizations. */
2050 struct regexp *const prog = (struct regexp *)SvANY(rx);
2051 /*register*/ char *s;
2052 register regnode *c;
2053 /*register*/ char *startpos = stringarg;
2054 I32 minlen; /* must match at least this many chars */
2055 I32 dontbother = 0; /* how many characters not to try at end */
2056 I32 end_shift = 0; /* Same for the end. */ /* CC */
2057 I32 scream_pos = -1; /* Internal iterator of scream. */
2058 char *scream_olds = NULL;
2059 const bool utf8_target = cBOOL(DO_UTF8(sv));
2061 RXi_GET_DECL(prog,progi);
2062 regmatch_info reginfo; /* create some info to pass to regtry etc */
2063 regexp_paren_pair *swap = NULL;
2064 GET_RE_DEBUG_FLAGS_DECL;
2066 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
2067 PERL_UNUSED_ARG(data);
2069 /* Be paranoid... */
2070 if (prog == NULL || startpos == NULL) {
2071 Perl_croak(aTHX_ "NULL regexp parameter");
2075 multiline = prog->extflags & RXf_PMf_MULTILINE;
2076 reginfo.prog = rx; /* Yes, sorry that this is confusing. */
2078 RX_MATCH_UTF8_set(rx, utf8_target);
2080 debug_start_match(rx, utf8_target, startpos, strend,
2084 minlen = prog->minlen;
2086 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
2087 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2088 "String too short [regexec_flags]...\n"));
2093 /* Check validity of program. */
2094 if (UCHARAT(progi->program) != REG_MAGIC) {
2095 Perl_croak(aTHX_ "corrupted regexp program");
2099 PL_reg_eval_set = 0;
2103 PL_reg_flags |= RF_utf8;
2105 /* Mark beginning of line for ^ and lookbehind. */
2106 reginfo.bol = startpos; /* XXX not used ??? */
2110 /* Mark end of line for $ (and such) */
2113 /* see how far we have to get to not match where we matched before */
2114 reginfo.till = startpos+minend;
2116 /* If there is a "must appear" string, look for it. */
2119 if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */
2121 if (flags & REXEC_IGNOREPOS){ /* Means: check only at start */
2122 reginfo.ganch = startpos + prog->gofs;
2123 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2124 "GPOS IGNOREPOS: reginfo.ganch = startpos + %"UVxf"\n",(UV)prog->gofs));
2125 } else if (sv && SvTYPE(sv) >= SVt_PVMG
2127 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
2128 && mg->mg_len >= 0) {
2129 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
2130 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2131 "GPOS MAGIC: reginfo.ganch = strbeg + %"IVdf"\n",(IV)mg->mg_len));
2133 if (prog->extflags & RXf_ANCH_GPOS) {
2134 if (s > reginfo.ganch)
2136 s = reginfo.ganch - prog->gofs;
2137 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2138 "GPOS ANCH_GPOS: s = ganch - %"UVxf"\n",(UV)prog->gofs));
2144 reginfo.ganch = strbeg + PTR2UV(data);
2145 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2146 "GPOS DATA: reginfo.ganch= strbeg + %"UVxf"\n",PTR2UV(data)));
2148 } else { /* pos() not defined */
2149 reginfo.ganch = strbeg;
2150 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2151 "GPOS: reginfo.ganch = strbeg\n"));
2154 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
2155 /* We have to be careful. If the previous successful match
2156 was from this regex we don't want a subsequent partially
2157 successful match to clobber the old results.
2158 So when we detect this possibility we add a swap buffer
2159 to the re, and switch the buffer each match. If we fail
2160 we switch it back, otherwise we leave it swapped.
2163 /* do we need a save destructor here for eval dies? */
2164 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
2166 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
2167 re_scream_pos_data d;
2169 d.scream_olds = &scream_olds;
2170 d.scream_pos = &scream_pos;
2171 s = re_intuit_start(rx, sv, s, strend, flags, &d);
2173 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
2174 goto phooey; /* not present */
2180 /* Simplest case: anchored match need be tried only once. */
2181 /* [unless only anchor is BOL and multiline is set] */
2182 if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
2183 if (s == startpos && regtry(®info, &startpos))
2185 else if (multiline || (prog->intflags & PREGf_IMPLICIT)
2186 || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
2191 dontbother = minlen - 1;
2192 end = HOP3c(strend, -dontbother, strbeg) - 1;
2193 /* for multiline we only have to try after newlines */
2194 if (prog->check_substr || prog->check_utf8) {
2195 /* because of the goto we can not easily reuse the macros for bifurcating the
2196 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2199 goto after_try_utf8;
2201 if (regtry(®info, &s)) {
2208 if (prog->extflags & RXf_USE_INTUIT) {
2209 s = re_intuit_start(rx, sv, s + UTF8SKIP(s), strend, flags, NULL);
2218 } /* end search for check string in unicode */
2220 if (s == startpos) {
2221 goto after_try_latin;
2224 if (regtry(®info, &s)) {
2231 if (prog->extflags & RXf_USE_INTUIT) {
2232 s = re_intuit_start(rx, sv, s + 1, strend, flags, NULL);
2241 } /* end search for check string in latin*/
2242 } /* end search for check string */
2243 else { /* search for newline */
2245 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
2248 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2249 while (s <= end) { /* note it could be possible to match at the end of the string */
2250 if (*s++ == '\n') { /* don't need PL_utf8skip here */
2251 if (regtry(®info, &s))
2255 } /* end search for newline */
2256 } /* end anchored/multiline check string search */
2258 } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
2260 /* the warning about reginfo.ganch being used without initialization
2261 is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN
2262 and we only enter this block when the same bit is set. */
2263 char *tmp_s = reginfo.ganch - prog->gofs;
2265 if (tmp_s >= strbeg && regtry(®info, &tmp_s))
2270 /* Messy cases: unanchored match. */
2271 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
2272 /* we have /x+whatever/ */
2273 /* it must be a one character string (XXXX Except UTF_PATTERN?) */
2278 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2279 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2280 ch = SvPVX_const(utf8_target ? prog->anchored_utf8 : prog->anchored_substr)[0];
2285 DEBUG_EXECUTE_r( did_match = 1 );
2286 if (regtry(®info, &s)) goto got_it;
2288 while (s < strend && *s == ch)
2296 DEBUG_EXECUTE_r( did_match = 1 );
2297 if (regtry(®info, &s)) goto got_it;
2299 while (s < strend && *s == ch)
2304 DEBUG_EXECUTE_r(if (!did_match)
2305 PerlIO_printf(Perl_debug_log,
2306 "Did not find anchored character...\n")
2309 else if (prog->anchored_substr != NULL
2310 || prog->anchored_utf8 != NULL
2311 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
2312 && prog->float_max_offset < strend - s)) {
2317 char *last1; /* Last position checked before */
2321 if (prog->anchored_substr || prog->anchored_utf8) {
2322 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2323 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2324 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
2325 back_max = back_min = prog->anchored_offset;
2327 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2328 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2329 must = utf8_target ? prog->float_utf8 : prog->float_substr;
2330 back_max = prog->float_max_offset;
2331 back_min = prog->float_min_offset;
2335 if (must == &PL_sv_undef)
2336 /* could not downgrade utf8 check substring, so must fail */
2342 last = HOP3c(strend, /* Cannot start after this */
2343 -(I32)(CHR_SVLEN(must)
2344 - (SvTAIL(must) != 0) + back_min), strbeg);
2347 last1 = HOPc(s, -1);
2349 last1 = s - 1; /* bogus */
2351 /* XXXX check_substr already used to find "s", can optimize if
2352 check_substr==must. */
2354 dontbother = end_shift;
2355 strend = HOPc(strend, -dontbother);
2356 while ( (s <= last) &&
2357 ((flags & REXEC_SCREAM) && SvSCREAM(sv)
2358 ? (s = screaminstr(sv, must, HOP3c(s, back_min, (back_min<0 ? strbeg : strend)) - strbeg,
2359 end_shift, &scream_pos, 0))
2360 : (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
2361 (unsigned char*)strend, must,
2362 multiline ? FBMrf_MULTILINE : 0))) ) {
2363 /* we may be pointing at the wrong string */
2364 if ((flags & REXEC_SCREAM) && RXp_MATCH_COPIED(prog))
2365 s = strbeg + (s - SvPVX_const(sv));
2366 DEBUG_EXECUTE_r( did_match = 1 );
2367 if (HOPc(s, -back_max) > last1) {
2368 last1 = HOPc(s, -back_min);
2369 s = HOPc(s, -back_max);
2372 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
2374 last1 = HOPc(s, -back_min);
2378 while (s <= last1) {
2379 if (regtry(®info, &s))
2385 while (s <= last1) {
2386 if (regtry(®info, &s))
2392 DEBUG_EXECUTE_r(if (!did_match) {
2393 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
2394 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2395 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2396 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2397 ? "anchored" : "floating"),
2398 quoted, RE_SV_TAIL(must));
2402 else if ( (c = progi->regstclass) ) {
2404 const OPCODE op = OP(progi->regstclass);
2405 /* don't bother with what can't match */
2406 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2407 strend = HOPc(strend, -(minlen - 1));
2410 SV * const prop = sv_newmortal();
2411 regprop(prog, prop, c);
2413 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
2415 PerlIO_printf(Perl_debug_log,
2416 "Matching stclass %.*s against %s (%d bytes)\n",
2417 (int)SvCUR(prop), SvPVX_const(prop),
2418 quoted, (int)(strend - s));
2421 if (find_byclass(prog, c, s, strend, ®info))
2423 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2427 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2432 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2433 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2434 float_real = utf8_target ? prog->float_utf8 : prog->float_substr;
2436 if ((flags & REXEC_SCREAM) && SvSCREAM(sv)) {
2437 last = screaminstr(sv, float_real, s - strbeg,
2438 end_shift, &scream_pos, 1); /* last one */
2440 last = scream_olds; /* Only one occurrence. */
2441 /* we may be pointing at the wrong string */
2442 else if (RXp_MATCH_COPIED(prog))
2443 s = strbeg + (s - SvPVX_const(sv));
2447 const char * const little = SvPV_const(float_real, len);
2449 if (SvTAIL(float_real)) {
2450 if (memEQ(strend - len + 1, little, len - 1))
2451 last = strend - len + 1;
2452 else if (!multiline)
2453 last = memEQ(strend - len, little, len)
2454 ? strend - len : NULL;
2460 last = rninstr(s, strend, little, little + len);
2462 last = strend; /* matching "$" */
2467 PerlIO_printf(Perl_debug_log,
2468 "%sCan't trim the tail, match fails (should not happen)%s\n",
2469 PL_colors[4], PL_colors[5]));
2470 goto phooey; /* Should not happen! */
2472 dontbother = strend - last + prog->float_min_offset;
2474 if (minlen && (dontbother < minlen))
2475 dontbother = minlen - 1;
2476 strend -= dontbother; /* this one's always in bytes! */
2477 /* We don't know much -- general case. */
2480 if (regtry(®info, &s))
2489 if (regtry(®info, &s))
2491 } while (s++ < strend);
2500 RX_MATCH_TAINTED_set(rx, PL_reg_flags & RF_tainted);
2502 if (PL_reg_eval_set)
2503 restore_pos(aTHX_ prog);
2504 if (RXp_PAREN_NAMES(prog))
2505 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
2507 /* make sure $`, $&, $', and $digit will work later */
2508 if ( !(flags & REXEC_NOT_FIRST) ) {
2509 RX_MATCH_COPY_FREE(rx);
2510 if (flags & REXEC_COPY_STR) {
2511 const I32 i = PL_regeol - startpos + (stringarg - strbeg);
2512 #ifdef PERL_OLD_COPY_ON_WRITE
2514 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2516 PerlIO_printf(Perl_debug_log,
2517 "Copy on write: regexp capture, type %d\n",
2520 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2521 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2522 assert (SvPOKp(prog->saved_copy));
2526 RX_MATCH_COPIED_on(rx);
2527 s = savepvn(strbeg, i);
2533 prog->subbeg = strbeg;
2534 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2541 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2542 PL_colors[4], PL_colors[5]));
2543 if (PL_reg_eval_set)
2544 restore_pos(aTHX_ prog);
2546 /* we failed :-( roll it back */
2547 Safefree(prog->offs);
2556 - regtry - try match at specific point
2558 STATIC I32 /* 0 failure, 1 success */
2559 S_regtry(pTHX_ regmatch_info *reginfo, char **startpos)
2563 REGEXP *const rx = reginfo->prog;
2564 regexp *const prog = (struct regexp *)SvANY(rx);
2565 RXi_GET_DECL(prog,progi);
2566 GET_RE_DEBUG_FLAGS_DECL;
2568 PERL_ARGS_ASSERT_REGTRY;
2570 reginfo->cutpoint=NULL;
2572 if ((prog->extflags & RXf_EVAL_SEEN) && !PL_reg_eval_set) {
2575 PL_reg_eval_set = RS_init;
2576 DEBUG_EXECUTE_r(DEBUG_s(
2577 PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n",
2578 (IV)(PL_stack_sp - PL_stack_base));
2581 cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base;
2582 /* Otherwise OP_NEXTSTATE will free whatever on stack now. */
2584 /* Apparently this is not needed, judging by wantarray. */
2585 /* SAVEI8(cxstack[cxstack_ix].blk_gimme);
2586 cxstack[cxstack_ix].blk_gimme = G_SCALAR; */
2589 /* Make $_ available to executed code. */
2590 if (reginfo->sv != DEFSV) {
2592 DEFSV_set(reginfo->sv);
2595 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2596 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2597 /* prepare for quick setting of pos */
2598 #ifdef PERL_OLD_COPY_ON_WRITE
2599 if (SvIsCOW(reginfo->sv))
2600 sv_force_normal_flags(reginfo->sv, 0);
2602 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2603 &PL_vtbl_mglob, NULL, 0);
2607 PL_reg_oldpos = mg->mg_len;
2608 SAVEDESTRUCTOR_X(restore_pos, prog);
2610 if (!PL_reg_curpm) {
2611 Newxz(PL_reg_curpm, 1, PMOP);
2614 SV* const repointer = &PL_sv_undef;
2615 /* this regexp is also owned by the new PL_reg_curpm, which
2616 will try to free it. */
2617 av_push(PL_regex_padav, repointer);
2618 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2619 PL_regex_pad = AvARRAY(PL_regex_padav);
2624 /* It seems that non-ithreads works both with and without this code.
2625 So for efficiency reasons it seems best not to have the code
2626 compiled when it is not needed. */
2627 /* This is safe against NULLs: */
2628 ReREFCNT_dec(PM_GETRE(PL_reg_curpm));
2629 /* PM_reg_curpm owns a reference to this regexp. */
2630 (void)ReREFCNT_inc(rx);
2632 PM_SETRE(PL_reg_curpm, rx);
2633 PL_reg_oldcurpm = PL_curpm;
2634 PL_curpm = PL_reg_curpm;
2635 if (RXp_MATCH_COPIED(prog)) {
2636 /* Here is a serious problem: we cannot rewrite subbeg,
2637 since it may be needed if this match fails. Thus
2638 $` inside (?{}) could fail... */
2639 PL_reg_oldsaved = prog->subbeg;
2640 PL_reg_oldsavedlen = prog->sublen;
2641 #ifdef PERL_OLD_COPY_ON_WRITE
2642 PL_nrs = prog->saved_copy;
2644 RXp_MATCH_COPIED_off(prog);
2647 PL_reg_oldsaved = NULL;
2648 prog->subbeg = PL_bostr;
2649 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2651 DEBUG_EXECUTE_r(PL_reg_starttry = *startpos);
2652 prog->offs[0].start = *startpos - PL_bostr;
2653 PL_reginput = *startpos;
2654 PL_reglastparen = &prog->lastparen;
2655 PL_reglastcloseparen = &prog->lastcloseparen;
2656 prog->lastparen = 0;
2657 prog->lastcloseparen = 0;
2659 PL_regoffs = prog->offs;
2660 if (PL_reg_start_tmpl <= prog->nparens) {
2661 PL_reg_start_tmpl = prog->nparens*3/2 + 3;
2662 if(PL_reg_start_tmp)
2663 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2665 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2668 /* XXXX What this code is doing here?!!! There should be no need
2669 to do this again and again, PL_reglastparen should take care of
2672 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2673 * Actually, the code in regcppop() (which Ilya may be meaning by
2674 * PL_reglastparen), is not needed at all by the test suite
2675 * (op/regexp, op/pat, op/split), but that code is needed otherwise
2676 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
2677 * Meanwhile, this code *is* needed for the
2678 * above-mentioned test suite tests to succeed. The common theme
2679 * on those tests seems to be returning null fields from matches.
2680 * --jhi updated by dapm */
2682 if (prog->nparens) {
2683 regexp_paren_pair *pp = PL_regoffs;
2685 for (i = prog->nparens; i > (I32)*PL_reglastparen; i--) {
2693 if (regmatch(reginfo, progi->program + 1)) {
2694 PL_regoffs[0].end = PL_reginput - PL_bostr;
2697 if (reginfo->cutpoint)
2698 *startpos= reginfo->cutpoint;
2699 REGCP_UNWIND(lastcp);
2704 #define sayYES goto yes
2705 #define sayNO goto no
2706 #define sayNO_SILENT goto no_silent
2708 /* we dont use STMT_START/END here because it leads to
2709 "unreachable code" warnings, which are bogus, but distracting. */
2710 #define CACHEsayNO \
2711 if (ST.cache_mask) \
2712 PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
2715 /* this is used to determine how far from the left messages like
2716 'failed...' are printed. It should be set such that messages
2717 are inline with the regop output that created them.
2719 #define REPORT_CODE_OFF 32
2722 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2723 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2725 #define SLAB_FIRST(s) (&(s)->states[0])
2726 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2728 /* grab a new slab and return the first slot in it */
2730 STATIC regmatch_state *
2733 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2736 regmatch_slab *s = PL_regmatch_slab->next;
2738 Newx(s, 1, regmatch_slab);
2739 s->prev = PL_regmatch_slab;
2741 PL_regmatch_slab->next = s;
2743 PL_regmatch_slab = s;
2744 return SLAB_FIRST(s);
2748 /* push a new state then goto it */
2750 #define PUSH_STATE_GOTO(state, node) \
2752 st->resume_state = state; \
2755 /* push a new state with success backtracking, then goto it */
2757 #define PUSH_YES_STATE_GOTO(state, node) \
2759 st->resume_state = state; \
2760 goto push_yes_state;
2766 regmatch() - main matching routine
2768 This is basically one big switch statement in a loop. We execute an op,
2769 set 'next' to point the next op, and continue. If we come to a point which
2770 we may need to backtrack to on failure such as (A|B|C), we push a
2771 backtrack state onto the backtrack stack. On failure, we pop the top
2772 state, and re-enter the loop at the state indicated. If there are no more
2773 states to pop, we return failure.
2775 Sometimes we also need to backtrack on success; for example /A+/, where
2776 after successfully matching one A, we need to go back and try to
2777 match another one; similarly for lookahead assertions: if the assertion
2778 completes successfully, we backtrack to the state just before the assertion
2779 and then carry on. In these cases, the pushed state is marked as
2780 'backtrack on success too'. This marking is in fact done by a chain of
2781 pointers, each pointing to the previous 'yes' state. On success, we pop to
2782 the nearest yes state, discarding any intermediate failure-only states.
2783 Sometimes a yes state is pushed just to force some cleanup code to be
2784 called at the end of a successful match or submatch; e.g. (??{$re}) uses
2785 it to free the inner regex.
2787 Note that failure backtracking rewinds the cursor position, while
2788 success backtracking leaves it alone.
2790 A pattern is complete when the END op is executed, while a subpattern
2791 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
2792 ops trigger the "pop to last yes state if any, otherwise return true"
2795 A common convention in this function is to use A and B to refer to the two
2796 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
2797 the subpattern to be matched possibly multiple times, while B is the entire
2798 rest of the pattern. Variable and state names reflect this convention.
2800 The states in the main switch are the union of ops and failure/success of
2801 substates associated with with that op. For example, IFMATCH is the op
2802 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
2803 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
2804 successfully matched A and IFMATCH_A_fail is a state saying that we have
2805 just failed to match A. Resume states always come in pairs. The backtrack
2806 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
2807 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
2808 on success or failure.
2810 The struct that holds a backtracking state is actually a big union, with
2811 one variant for each major type of op. The variable st points to the
2812 top-most backtrack struct. To make the code clearer, within each
2813 block of code we #define ST to alias the relevant union.
2815 Here's a concrete example of a (vastly oversimplified) IFMATCH
2821 #define ST st->u.ifmatch
2823 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2824 ST.foo = ...; // some state we wish to save
2826 // push a yes backtrack state with a resume value of
2827 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
2829 PUSH_YES_STATE_GOTO(IFMATCH_A, A);
2832 case IFMATCH_A: // we have successfully executed A; now continue with B
2834 bar = ST.foo; // do something with the preserved value
2837 case IFMATCH_A_fail: // A failed, so the assertion failed
2838 ...; // do some housekeeping, then ...
2839 sayNO; // propagate the failure
2846 For any old-timers reading this who are familiar with the old recursive
2847 approach, the code above is equivalent to:
2849 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2858 ...; // do some housekeeping, then ...
2859 sayNO; // propagate the failure
2862 The topmost backtrack state, pointed to by st, is usually free. If you
2863 want to claim it, populate any ST.foo fields in it with values you wish to
2864 save, then do one of
2866 PUSH_STATE_GOTO(resume_state, node);
2867 PUSH_YES_STATE_GOTO(resume_state, node);
2869 which sets that backtrack state's resume value to 'resume_state', pushes a
2870 new free entry to the top of the backtrack stack, then goes to 'node'.
2871 On backtracking, the free slot is popped, and the saved state becomes the
2872 new free state. An ST.foo field in this new top state can be temporarily
2873 accessed to retrieve values, but once the main loop is re-entered, it
2874 becomes available for reuse.
2876 Note that the depth of the backtrack stack constantly increases during the
2877 left-to-right execution of the pattern, rather than going up and down with
2878 the pattern nesting. For example the stack is at its maximum at Z at the
2879 end of the pattern, rather than at X in the following:
2881 /(((X)+)+)+....(Y)+....Z/
2883 The only exceptions to this are lookahead/behind assertions and the cut,
2884 (?>A), which pop all the backtrack states associated with A before
2887 Backtrack state structs are allocated in slabs of about 4K in size.
2888 PL_regmatch_state and st always point to the currently active state,
2889 and PL_regmatch_slab points to the slab currently containing
2890 PL_regmatch_state. The first time regmatch() is called, the first slab is
2891 allocated, and is never freed until interpreter destruction. When the slab
2892 is full, a new one is allocated and chained to the end. At exit from
2893 regmatch(), slabs allocated since entry are freed.
2898 #define DEBUG_STATE_pp(pp) \
2900 DUMP_EXEC_POS(locinput, scan, utf8_target); \
2901 PerlIO_printf(Perl_debug_log, \
2902 " %*s"pp" %s%s%s%s%s\n", \
2904 PL_reg_name[st->resume_state], \
2905 ((st==yes_state||st==mark_state) ? "[" : ""), \
2906 ((st==yes_state) ? "Y" : ""), \
2907 ((st==mark_state) ? "M" : ""), \
2908 ((st==yes_state||st==mark_state) ? "]" : "") \
2913 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
2918 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
2919 const char *start, const char *end, const char *blurb)
2921 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
2923 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
2928 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
2929 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
2931 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
2932 start, end - start, 60);
2934 PerlIO_printf(Perl_debug_log,
2935 "%s%s REx%s %s against %s\n",
2936 PL_colors[4], blurb, PL_colors[5], s0, s1);
2938 if (utf8_target||utf8_pat)
2939 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
2940 utf8_pat ? "pattern" : "",
2941 utf8_pat && utf8_target ? " and " : "",
2942 utf8_target ? "string" : ""
2948 S_dump_exec_pos(pTHX_ const char *locinput,
2949 const regnode *scan,
2950 const char *loc_regeol,
2951 const char *loc_bostr,
2952 const char *loc_reg_starttry,
2953 const bool utf8_target)
2955 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
2956 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
2957 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
2958 /* The part of the string before starttry has one color
2959 (pref0_len chars), between starttry and current
2960 position another one (pref_len - pref0_len chars),
2961 after the current position the third one.
2962 We assume that pref0_len <= pref_len, otherwise we
2963 decrease pref0_len. */
2964 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
2965 ? (5 + taill) - l : locinput - loc_bostr;
2968 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
2970 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
2972 pref0_len = pref_len - (locinput - loc_reg_starttry);
2973 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
2974 l = ( loc_regeol - locinput > (5 + taill) - pref_len
2975 ? (5 + taill) - pref_len : loc_regeol - locinput);
2976 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
2980 if (pref0_len > pref_len)
2981 pref0_len = pref_len;
2983 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
2985 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
2986 (locinput - pref_len),pref0_len, 60, 4, 5);
2988 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
2989 (locinput - pref_len + pref0_len),
2990 pref_len - pref0_len, 60, 2, 3);
2992 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
2993 locinput, loc_regeol - locinput, 10, 0, 1);
2995 const STRLEN tlen=len0+len1+len2;
2996 PerlIO_printf(Perl_debug_log,
2997 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
2998 (IV)(locinput - loc_bostr),
3001 (docolor ? "" : "> <"),
3003 (int)(tlen > 19 ? 0 : 19 - tlen),
3010 /* reg_check_named_buff_matched()
3011 * Checks to see if a named buffer has matched. The data array of
3012 * buffer numbers corresponding to the buffer is expected to reside
3013 * in the regexp->data->data array in the slot stored in the ARG() of
3014 * node involved. Note that this routine doesn't actually care about the
3015 * name, that information is not preserved from compilation to execution.
3016 * Returns the index of the leftmost defined buffer with the given name
3017 * or 0 if non of the buffers matched.
3020 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan)
3023 RXi_GET_DECL(rex,rexi);
3024 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
3025 I32 *nums=(I32*)SvPVX(sv_dat);
3027 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
3029 for ( n=0; n<SvIVX(sv_dat); n++ ) {
3030 if ((I32)*PL_reglastparen >= nums[n] &&
3031 PL_regoffs[nums[n]].end != -1)
3040 /* free all slabs above current one - called during LEAVE_SCOPE */
3043 S_clear_backtrack_stack(pTHX_ void *p)
3045 regmatch_slab *s = PL_regmatch_slab->next;
3050 PL_regmatch_slab->next = NULL;
3052 regmatch_slab * const osl = s;
3059 #define SETREX(Re1,Re2) \
3060 if (PL_reg_eval_set) PM_SETRE((PL_reg_curpm), (Re2)); \
3063 STATIC I32 /* 0 failure, 1 success */
3064 S_regmatch(pTHX_ regmatch_info *reginfo, regnode *prog)
3066 #if PERL_VERSION < 9 && !defined(PERL_CORE)
3070 register const bool utf8_target = PL_reg_match_utf8;
3071 const U32 uniflags = UTF8_ALLOW_DEFAULT;
3072 REGEXP *rex_sv = reginfo->prog;
3073 regexp *rex = (struct regexp *)SvANY(rex_sv);
3074 RXi_GET_DECL(rex,rexi);
3076 /* the current state. This is a cached copy of PL_regmatch_state */
3077 register regmatch_state *st;
3078 /* cache heavy used fields of st in registers */
3079 register regnode *scan;
3080 register regnode *next;
3081 register U32 n = 0; /* general value; init to avoid compiler warning */
3082 register I32 ln = 0; /* len or last; init to avoid compiler warning */
3083 register char *locinput = PL_reginput;
3084 register I32 nextchr; /* is always set to UCHARAT(locinput) */
3086 bool result = 0; /* return value of S_regmatch */
3087 int depth = 0; /* depth of backtrack stack */
3088 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
3089 const U32 max_nochange_depth =
3090 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
3091 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
3092 regmatch_state *yes_state = NULL; /* state to pop to on success of
3094 /* mark_state piggy backs on the yes_state logic so that when we unwind
3095 the stack on success we can update the mark_state as we go */
3096 regmatch_state *mark_state = NULL; /* last mark state we have seen */
3097 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
3098 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
3100 bool no_final = 0; /* prevent failure from backtracking? */
3101 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
3102 char *startpoint = PL_reginput;
3103 SV *popmark = NULL; /* are we looking for a mark? */
3104 SV *sv_commit = NULL; /* last mark name seen in failure */
3105 SV *sv_yes_mark = NULL; /* last mark name we have seen
3106 during a successful match */
3107 U32 lastopen = 0; /* last open we saw */
3108 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
3109 SV* const oreplsv = GvSV(PL_replgv);
3110 /* these three flags are set by various ops to signal information to
3111 * the very next op. They have a useful lifetime of exactly one loop
3112 * iteration, and are not preserved or restored by state pushes/pops
3114 bool sw = 0; /* the condition value in (?(cond)a|b) */
3115 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
3116 int logical = 0; /* the following EVAL is:
3120 or the following IFMATCH/UNLESSM is:
3121 false: plain (?=foo)
3122 true: used as a condition: (?(?=foo))
3125 GET_RE_DEBUG_FLAGS_DECL;
3128 PERL_ARGS_ASSERT_REGMATCH;
3130 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
3131 PerlIO_printf(Perl_debug_log,"regmatch start\n");
3133 /* on first ever call to regmatch, allocate first slab */
3134 if (!PL_regmatch_slab) {
3135 Newx(PL_regmatch_slab, 1, regmatch_slab);
3136 PL_regmatch_slab->prev = NULL;
3137 PL_regmatch_slab->next = NULL;
3138 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
3141 oldsave = PL_savestack_ix;
3142 SAVEDESTRUCTOR_X(S_clear_backtrack_stack, NULL);
3143 SAVEVPTR(PL_regmatch_slab);
3144 SAVEVPTR(PL_regmatch_state);
3146 /* grab next free state slot */
3147 st = ++PL_regmatch_state;
3148 if (st > SLAB_LAST(PL_regmatch_slab))
3149 st = PL_regmatch_state = S_push_slab(aTHX);
3151 /* Note that nextchr is a byte even in UTF */
3152 nextchr = UCHARAT(locinput);
3154 while (scan != NULL) {
3157 SV * const prop = sv_newmortal();
3158 regnode *rnext=regnext(scan);
3159 DUMP_EXEC_POS( locinput, scan, utf8_target );
3160 regprop(rex, prop, scan);
3162 PerlIO_printf(Perl_debug_log,
3163 "%3"IVdf":%*s%s(%"IVdf")\n",
3164 (IV)(scan - rexi->program), depth*2, "",
3166 (PL_regkind[OP(scan)] == END || !rnext) ?
3167 0 : (IV)(rnext - rexi->program));
3170 next = scan + NEXT_OFF(scan);
3173 state_num = OP(scan);
3177 assert(PL_reglastparen == &rex->lastparen);
3178 assert(PL_reglastcloseparen == &rex->lastcloseparen);
3179 assert(PL_regoffs == rex->offs);
3181 switch (state_num) {
3183 if (locinput == PL_bostr)
3185 /* reginfo->till = reginfo->bol; */
3190 if (locinput == PL_bostr ||
3191 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
3197 if (locinput == PL_bostr)
3201 if (locinput == reginfo->ganch)
3206 /* update the startpoint */
3207 st->u.keeper.val = PL_regoffs[0].start;
3208 PL_reginput = locinput;
3209 PL_regoffs[0].start = locinput - PL_bostr;
3210 PUSH_STATE_GOTO(KEEPS_next, next);
3212 case KEEPS_next_fail:
3213 /* rollback the start point change */
3214 PL_regoffs[0].start = st->u.keeper.val;
3220 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3225 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3227 if (PL_regeol - locinput > 1)
3231 if (PL_regeol != locinput)
3235 if (!nextchr && locinput >= PL_regeol)
3238 locinput += PL_utf8skip[nextchr];
3239 if (locinput > PL_regeol)
3241 nextchr = UCHARAT(locinput);
3244 nextchr = UCHARAT(++locinput);
3247 if (!nextchr && locinput >= PL_regeol)
3249 nextchr = UCHARAT(++locinput);
3252 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
3255 locinput += PL_utf8skip[nextchr];
3256 if (locinput > PL_regeol)
3258 nextchr = UCHARAT(locinput);
3261 nextchr = UCHARAT(++locinput);
3265 #define ST st->u.trie
3267 /* In this case the charclass data is available inline so
3268 we can fail fast without a lot of extra overhead.
3270 if (scan->flags == EXACT || !utf8_target) {
3271 if(!ANYOF_BITMAP_TEST(scan, *locinput)) {
3273 PerlIO_printf(Perl_debug_log,
3274 "%*s %sfailed to match trie start class...%s\n",
3275 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3283 /* the basic plan of execution of the trie is:
3284 * At the beginning, run though all the states, and
3285 * find the longest-matching word. Also remember the position
3286 * of the shortest matching word. For example, this pattern:
3289 * when matched against the string "abcde", will generate
3290 * accept states for all words except 3, with the longest
3291 * matching word being 4, and the shortest being 1 (with
3292 * the position being after char 1 of the string).
3294 * Then for each matching word, in word order (i.e. 1,2,4,5),
3295 * we run the remainder of the pattern; on each try setting
3296 * the current position to the character following the word,
3297 * returning to try the next word on failure.
3299 * We avoid having to build a list of words at runtime by
3300 * using a compile-time structure, wordinfo[].prev, which
3301 * gives, for each word, the previous accepting word (if any).
3302 * In the case above it would contain the mappings 1->2, 2->0,
3303 * 3->0, 4->5, 5->1. We can use this table to generate, from
3304 * the longest word (4 above), a list of all words, by
3305 * following the list of prev pointers; this gives us the
3306 * unordered list 4,5,1,2. Then given the current word we have
3307 * just tried, we can go through the list and find the
3308 * next-biggest word to try (so if we just failed on word 2,
3309 * the next in the list is 4).
3311 * Since at runtime we don't record the matching position in
3312 * the string for each word, we have to work that out for
3313 * each word we're about to process. The wordinfo table holds
3314 * the character length of each word; given that we recorded
3315 * at the start: the position of the shortest word and its
3316 * length in chars, we just need to move the pointer the
3317 * difference between the two char lengths. Depending on
3318 * Unicode status and folding, that's cheap or expensive.
3320 * This algorithm is optimised for the case where are only a
3321 * small number of accept states, i.e. 0,1, or maybe 2.
3322 * With lots of accepts states, and having to try all of them,
3323 * it becomes quadratic on number of accept states to find all
3328 /* what type of TRIE am I? (utf8 makes this contextual) */
3329 DECL_TRIE_TYPE(scan);
3331 /* what trie are we using right now */
3332 reg_trie_data * const trie
3333 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
3334 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
3335 U32 state = trie->startstate;
3337 if (trie->bitmap && trie_type != trie_utf8_fold &&
3338 !TRIE_BITMAP_TEST(trie,*locinput)
3340 if (trie->states[ state ].wordnum) {
3342 PerlIO_printf(Perl_debug_log,
3343 "%*s %smatched empty string...%s\n",
3344 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3350 PerlIO_printf(Perl_debug_log,
3351 "%*s %sfailed to match trie start class...%s\n",
3352 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3359 U8 *uc = ( U8* )locinput;
3363 U8 *uscan = (U8*)NULL;
3364 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
3365 U32 charcount = 0; /* how many input chars we have matched */
3366 U32 accepted = 0; /* have we seen any accepting states? */
3369 ST.jump = trie->jump;
3372 ST.longfold = FALSE; /* char longer if folded => it's harder */
3375 /* fully traverse the TRIE; note the position of the
3376 shortest accept state and the wordnum of the longest
3379 while ( state && uc <= (U8*)PL_regeol ) {
3380 U32 base = trie->states[ state ].trans.base;
3384 wordnum = trie->states[ state ].wordnum;
3386 if (wordnum) { /* it's an accept state */
3389 /* record first match position */
3391 ST.firstpos = (U8*)locinput;
3396 ST.firstchars = charcount;
3399 if (!ST.nextword || wordnum < ST.nextword)
3400 ST.nextword = wordnum;
3401 ST.topword = wordnum;
3404 DEBUG_TRIE_EXECUTE_r({
3405 DUMP_EXEC_POS( (char *)uc, scan, utf8_target );
3406 PerlIO_printf( Perl_debug_log,
3407 "%*s %sState: %4"UVxf" Accepted: %c ",
3408 2+depth * 2, "", PL_colors[4],
3409 (UV)state, (accepted ? 'Y' : 'N'));
3412 /* read a char and goto next state */
3415 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
3416 uscan, len, uvc, charid, foldlen,
3423 base + charid - 1 - trie->uniquecharcount)) >= 0)
3425 && ((U32)offset < trie->lasttrans)
3426 && trie->trans[offset].check == state)
3428 state = trie->trans[offset].next;
3439 DEBUG_TRIE_EXECUTE_r(
3440 PerlIO_printf( Perl_debug_log,
3441 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
3442 charid, uvc, (UV)state, PL_colors[5] );
3448 /* calculate total number of accept states */
3453 w = trie->wordinfo[w].prev;
3456 ST.accepted = accepted;
3460 PerlIO_printf( Perl_debug_log,
3461 "%*s %sgot %"IVdf" possible matches%s\n",
3462 REPORT_CODE_OFF + depth * 2, "",
3463 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
3465 goto trie_first_try; /* jump into the fail handler */
3469 case TRIE_next_fail: /* we failed - try next alternative */
3471 REGCP_UNWIND(ST.cp);
3472 for (n = *PL_reglastparen; n > ST.lastparen; n--)
3473 PL_regoffs[n].end = -1;
3474 *PL_reglastparen = n;
3476 if (!--ST.accepted) {
3478 PerlIO_printf( Perl_debug_log,
3479 "%*s %sTRIE failed...%s\n",
3480 REPORT_CODE_OFF+depth*2, "",
3487 /* Find next-highest word to process. Note that this code
3488 * is O(N^2) per trie run (O(N) per branch), so keep tight */
3489 register U16 min = 0;
3491 register U16 const nextword = ST.nextword;
3492 register reg_trie_wordinfo * const wordinfo
3493 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
3494 for (word=ST.topword; word; word=wordinfo[word].prev) {
3495 if (word > nextword && (!min || word < min))
3508 ST.lastparen = *PL_reglastparen;
3512 /* find start char of end of current word */
3514 U32 chars; /* how many chars to skip */
3515 U8 *uc = ST.firstpos;
3516 reg_trie_data * const trie
3517 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
3519 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
3521 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
3525 /* the hard option - fold each char in turn and find
3526 * its folded length (which may be different */
3527 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
3535 uvc = utf8n_to_uvuni((U8*)uc, UTF8_MAXLEN, &len,
3543 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
3548 uvc = utf8n_to_uvuni(uscan, UTF8_MAXLEN, &len,
3562 PL_reginput = (char *)uc;
3565 scan = (ST.jump && ST.jump[ST.nextword])
3566 ? ST.me + ST.jump[ST.nextword]
3570 PerlIO_printf( Perl_debug_log,
3571 "%*s %sTRIE matched word #%d, continuing%s\n",
3572 REPORT_CODE_OFF+depth*2, "",
3579 if (ST.accepted > 1 || has_cutgroup) {
3580 PUSH_STATE_GOTO(TRIE_next, scan);
3583 /* only one choice left - just continue */
3585 AV *const trie_words
3586 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
3587 SV ** const tmp = av_fetch( trie_words,
3589 SV *sv= tmp ? sv_newmortal() : NULL;
3591 PerlIO_printf( Perl_debug_log,
3592 "%*s %sonly one match left, short-circuiting: #%d <%s>%s\n",
3593 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
3595 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
3596 PL_colors[0], PL_colors[1],
3597 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII
3599 : "not compiled under -Dr",
3603 locinput = PL_reginput;
3604 nextchr = UCHARAT(locinput);
3605 continue; /* execute rest of RE */
3610 char *s = STRING(scan);
3612 if (utf8_target != UTF_PATTERN) {
3613 /* The target and the pattern have differing utf8ness. */
3615 const char * const e = s + ln;
3618 /* The target is utf8, the pattern is not utf8. */
3623 if (NATIVE_TO_UNI(*(U8*)s) !=
3624 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
3632 /* The target is not utf8, the pattern is utf8. */
3637 if (NATIVE_TO_UNI(*((U8*)l)) !=
3638 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
3646 nextchr = UCHARAT(locinput);
3649 /* The target and the pattern have the same utf8ness. */
3650 /* Inline the first character, for speed. */
3651 if (UCHARAT(s) != nextchr)
3653 if (PL_regeol - locinput < ln)
3655 if (ln > 1 && memNE(s, locinput, ln))
3658 nextchr = UCHARAT(locinput);
3663 const U8 * fold_array;
3665 U32 fold_utf8_flags;
3667 PL_reg_flags |= RF_tainted;
3668 folder = foldEQ_locale;
3669 fold_array = PL_fold_locale;
3670 fold_utf8_flags = FOLDEQ_UTF8_LOCALE;
3674 case EXACTFU_NO_TRIE:
3676 folder = foldEQ_latin1;
3677 fold_array = PL_fold_latin1;
3678 fold_utf8_flags = (UTF_PATTERN) ? FOLDEQ_S1_ALREADY_FOLDED : 0;
3682 folder = foldEQ_latin1;
3683 fold_array = PL_fold_latin1;
3684 fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
3689 fold_array = PL_fold;
3690 fold_utf8_flags = 0;
3696 if (utf8_target || UTF_PATTERN || state_num == EXACTFU_SS) {
3697 /* Either target or the pattern are utf8, or has the issue where
3698 * the fold lengths may differ. */
3699 const char * const l = locinput;
3700 char *e = PL_regeol;
3702 if (! foldEQ_utf8_flags(s, 0, ln, cBOOL(UTF_PATTERN),
3703 l, &e, 0, utf8_target, fold_utf8_flags))
3708 nextchr = UCHARAT(locinput);
3712 /* Neither the target nor the pattern are utf8 */
3713 if (UCHARAT(s) != nextchr &&
3714 UCHARAT(s) != fold_array[nextchr])
3718 if (PL_regeol - locinput < ln)
3720 if (ln > 1 && ! folder(s, locinput, ln))
3723 nextchr = UCHARAT(locinput);
3727 /* XXX Could improve efficiency by separating these all out using a
3728 * macro or in-line function. At that point regcomp.c would no longer
3729 * have to set the FLAGS fields of these */
3732 PL_reg_flags |= RF_tainted;
3740 /* was last char in word? */
3742 && FLAGS(scan) != REGEX_ASCII_RESTRICTED_CHARSET
3743 && FLAGS(scan) != REGEX_ASCII_MORE_RESTRICTED_CHARSET)
3745 if (locinput == PL_bostr)
3748 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
3750 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
3752 if (FLAGS(scan) != REGEX_LOCALE_CHARSET) {
3753 ln = isALNUM_uni(ln);
3754 LOAD_UTF8_CHARCLASS_ALNUM();
3755 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, utf8_target);
3758 ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
3759 n = isALNUM_LC_utf8((U8*)locinput);
3764 /* Here the string isn't utf8, or is utf8 and only ascii
3765 * characters are to match \w. In the latter case looking at
3766 * the byte just prior to the current one may be just the final
3767 * byte of a multi-byte character. This is ok. There are two
3769 * 1) it is a single byte character, and then the test is doing
3770 * just what it's supposed to.
3771 * 2) it is a multi-byte character, in which case the final
3772 * byte is never mistakable for ASCII, and so the test
3773 * will say it is not a word character, which is the
3774 * correct answer. */
3775 ln = (locinput != PL_bostr) ?
3776 UCHARAT(locinput - 1) : '\n';
3777 switch (FLAGS(scan)) {
3778 case REGEX_UNICODE_CHARSET:
3779 ln = isWORDCHAR_L1(ln);
3780 n = isWORDCHAR_L1(nextchr);
3782 case REGEX_LOCALE_CHARSET:
3783 ln = isALNUM_LC(ln);
3784 n = isALNUM_LC(nextchr);
3786 case REGEX_DEPENDS_CHARSET:
3788 n = isALNUM(nextchr);
3790 case REGEX_ASCII_RESTRICTED_CHARSET:
3791 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
3792 ln = isWORDCHAR_A(ln);
3793 n = isWORDCHAR_A(nextchr);
3796 Perl_croak(aTHX_ "panic: Unexpected FLAGS %u in op %u", FLAGS(scan), OP(scan));
3800 /* Note requires that all BOUNDs be lower than all NBOUNDs in
3802 if (((!ln) == (!n)) == (OP(scan) < NBOUND))
3807 if (utf8_target || state_num == ANYOFV) {
3808 STRLEN inclasslen = PL_regeol - locinput;
3809 if (locinput >= PL_regeol)
3812 if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, utf8_target))
3814 locinput += inclasslen;
3815 nextchr = UCHARAT(locinput);
3820 nextchr = UCHARAT(locinput);
3821 if (!nextchr && locinput >= PL_regeol)
3823 if (!REGINCLASS(rex, scan, (U8*)locinput))
3825 nextchr = UCHARAT(++locinput);
3829 /* Special char classes - The defines start on line 129 or so */
3830 CCC_TRY_U(ALNUM, NALNUM, isWORDCHAR,
3831 ALNUML, NALNUML, isALNUM_LC, isALNUM_LC_utf8,
3832 ALNUMU, NALNUMU, isWORDCHAR_L1,
3833 ALNUMA, NALNUMA, isWORDCHAR_A,
3836 CCC_TRY_U(SPACE, NSPACE, isSPACE,
3837 SPACEL, NSPACEL, isSPACE_LC, isSPACE_LC_utf8,
3838 SPACEU, NSPACEU, isSPACE_L1,
3839 SPACEA, NSPACEA, isSPACE_A,
3842 CCC_TRY(DIGIT, NDIGIT, isDIGIT,
3843 DIGITL, NDIGITL, isDIGIT_LC, isDIGIT_LC_utf8,
3844 DIGITA, NDIGITA, isDIGIT_A,
3847 case CLUMP: /* Match \X: logical Unicode character. This is defined as
3848 a Unicode extended Grapheme Cluster */
3849 /* From http://www.unicode.org/reports/tr29 (5.2 version). An
3850 extended Grapheme Cluster is:
3853 | Prepend* Begin Extend*
3856 Begin is (Hangul-syllable | ! Control)
3857 Extend is (Grapheme_Extend | Spacing_Mark)
3858 Control is [ GCB_Control CR LF ]
3860 The discussion below shows how the code for CLUMP is derived
3861 from this regex. Note that most of these concepts are from
3862 property values of the Grapheme Cluster Boundary (GCB) property.
3863 No code point can have multiple property values for a given
3864 property. Thus a code point in Prepend can't be in Control, but
3865 it must be in !Control. This is why Control above includes
3866 GCB_Control plus CR plus LF. The latter two are used in the GCB
3867 property separately, and so can't be in GCB_Control, even though
3868 they logically are controls. Control is not the same as gc=cc,
3869 but includes format and other characters as well.
3871 The Unicode definition of Hangul-syllable is:
3873 | (L* ( ( V | LV ) V* | LVT ) T*)
3876 Each of these is a value for the GCB property, and hence must be
3877 disjoint, so the order they are tested is immaterial, so the
3878 above can safely be changed to
3881 | (L* ( LVT | ( V | LV ) V*) T*)
3883 The last two terms can be combined like this:
3885 | (( LVT | ( V | LV ) V*) T*))
3887 And refactored into this:
3888 L* (L | LVT T* | V V* T* | LV V* T*)
3890 That means that if we have seen any L's at all we can quit
3891 there, but if the next character is an LVT, a V, or an LV we
3894 There is a subtlety with Prepend* which showed up in testing.
3895 Note that the Begin, and only the Begin is required in:
3896 | Prepend* Begin Extend*
3897 Also, Begin contains '! Control'. A Prepend must be a
3898 '! Control', which means it must also be a Begin. What it
3899 comes down to is that if we match Prepend* and then find no
3900 suitable Begin afterwards, that if we backtrack the last
3901 Prepend, that one will be a suitable Begin.
3904 if (locinput >= PL_regeol)
3906 if (! utf8_target) {
3908 /* Match either CR LF or '.', as all the other possibilities
3910 locinput++; /* Match the . or CR */
3911 if (nextchr == '\r' /* And if it was CR, and the next is LF,
3913 && locinput < PL_regeol
3914 && UCHARAT(locinput) == '\n') locinput++;
3918 /* Utf8: See if is ( CR LF ); already know that locinput <
3919 * PL_regeol, so locinput+1 is in bounds */
3920 if (nextchr == '\r' && UCHARAT(locinput + 1) == '\n') {
3924 /* In case have to backtrack to beginning, then match '.' */
3925 char *starting = locinput;
3927 /* In case have to backtrack the last prepend */
3928 char *previous_prepend = 0;
3930 LOAD_UTF8_CHARCLASS_GCB();
3932 /* Match (prepend)* */
3933 while (locinput < PL_regeol
3934 && swash_fetch(PL_utf8_X_prepend,
3935 (U8*)locinput, utf8_target))
3937 previous_prepend = locinput;
3938 locinput += UTF8SKIP(locinput);
3941 /* As noted above, if we matched a prepend character, but
3942 * the next thing won't match, back off the last prepend we
3943 * matched, as it is guaranteed to match the begin */
3944 if (previous_prepend
3945 && (locinput >= PL_regeol
3946 || ! swash_fetch(PL_utf8_X_begin,
3947 (U8*)locinput, utf8_target)))
3949 locinput = previous_prepend;
3952 /* Note that here we know PL_regeol > locinput, as we
3953 * tested that upon input to this switch case, and if we
3954 * moved locinput forward, we tested the result just above
3955 * and it either passed, or we backed off so that it will
3957 if (! swash_fetch(PL_utf8_X_begin, (U8*)locinput, utf8_target)) {
3959 /* Here did not match the required 'Begin' in the
3960 * second term. So just match the very first
3961 * character, the '.' of the final term of the regex */
3962 locinput = starting + UTF8SKIP(starting);
3965 /* Here is the beginning of a character that can have
3966 * an extender. It is either a hangul syllable, or a
3968 if (swash_fetch(PL_utf8_X_non_hangul,
3969 (U8*)locinput, utf8_target))
3972 /* Here not a Hangul syllable, must be a
3973 * ('! * Control') */
3974 locinput += UTF8SKIP(locinput);
3977 /* Here is a Hangul syllable. It can be composed
3978 * of several individual characters. One
3979 * possibility is T+ */
3980 if (swash_fetch(PL_utf8_X_T,
3981 (U8*)locinput, utf8_target))
3983 while (locinput < PL_regeol
3984 && swash_fetch(PL_utf8_X_T,
3985 (U8*)locinput, utf8_target))
3987 locinput += UTF8SKIP(locinput);
3991 /* Here, not T+, but is a Hangul. That means
3992 * it is one of the others: L, LV, LVT or V,
3994 * L* (L | LVT T* | V V* T* | LV V* T*) */
3997 while (locinput < PL_regeol
3998 && swash_fetch(PL_utf8_X_L,
3999 (U8*)locinput, utf8_target))
4001 locinput += UTF8SKIP(locinput);
4004 /* Here, have exhausted L*. If the next
4005 * character is not an LV, LVT nor V, it means
4006 * we had to have at least one L, so matches L+
4007 * in the original equation, we have a complete
4008 * hangul syllable. Are done. */
4010 if (locinput < PL_regeol
4011 && swash_fetch(PL_utf8_X_LV_LVT_V,
4012 (U8*)locinput, utf8_target))
4015 /* Otherwise keep going. Must be LV, LVT
4016 * or V. See if LVT */
4017 if (swash_fetch(PL_utf8_X_LVT,
4018 (U8*)locinput, utf8_target))
4020 locinput += UTF8SKIP(locinput);
4023 /* Must be V or LV. Take it, then
4025 locinput += UTF8SKIP(locinput);
4026 while (locinput < PL_regeol
4027 && swash_fetch(PL_utf8_X_V,
4028 (U8*)locinput, utf8_target))
4030 locinput += UTF8SKIP(locinput);
4034 /* And any of LV, LVT, or V can be followed
4036 while (locinput < PL_regeol
4037 && swash_fetch(PL_utf8_X_T,
4041 locinput += UTF8SKIP(locinput);
4047 /* Match any extender */
4048 while (locinput < PL_regeol
4049 && swash_fetch(PL_utf8_X_extend,
4050 (U8*)locinput, utf8_target))
4052 locinput += UTF8SKIP(locinput);
4056 if (locinput > PL_regeol) sayNO;
4058 nextchr = UCHARAT(locinput);
4062 { /* The capture buffer cases. The ones beginning with N for the
4063 named buffers just convert to the equivalent numbered and
4064 pretend they were called as the corresponding numbered buffer
4066 /* don't initialize these in the declaration, it makes C++
4071 const U8 *fold_array;
4074 PL_reg_flags |= RF_tainted;
4075 folder = foldEQ_locale;
4076 fold_array = PL_fold_locale;
4078 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4082 folder = foldEQ_latin1;
4083 fold_array = PL_fold_latin1;
4085 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4089 folder = foldEQ_latin1;
4090 fold_array = PL_fold_latin1;
4092 utf8_fold_flags = 0;
4097 fold_array = PL_fold;
4099 utf8_fold_flags = 0;
4106 utf8_fold_flags = 0;
4109 /* For the named back references, find the corresponding buffer
4111 n = reg_check_named_buff_matched(rex,scan);
4116 goto do_nref_ref_common;
4119 PL_reg_flags |= RF_tainted;
4120 folder = foldEQ_locale;
4121 fold_array = PL_fold_locale;
4122 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4126 folder = foldEQ_latin1;
4127 fold_array = PL_fold_latin1;
4128 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4132 folder = foldEQ_latin1;
4133 fold_array = PL_fold_latin1;
4134 utf8_fold_flags = 0;
4139 fold_array = PL_fold;
4140 utf8_fold_flags = 0;
4146 utf8_fold_flags = 0;
4150 n = ARG(scan); /* which paren pair */
4153 ln = PL_regoffs[n].start;
4154 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
4155 if (*PL_reglastparen < n || ln == -1)
4156 sayNO; /* Do not match unless seen CLOSEn. */
4157 if (ln == PL_regoffs[n].end)
4161 if (type != REF /* REF can do byte comparison */
4162 && (utf8_target || type == REFFU))
4163 { /* XXX handle REFFL better */
4164 char * limit = PL_regeol;
4166 /* This call case insensitively compares the entire buffer
4167 * at s, with the current input starting at locinput, but
4168 * not going off the end given by PL_regeol, and returns in
4169 * limit upon success, how much of the current input was
4171 if (! foldEQ_utf8_flags(s, NULL, PL_regoffs[n].end - ln, utf8_target,
4172 locinput, &limit, 0, utf8_target, utf8_fold_flags))
4177 nextchr = UCHARAT(locinput);
4181 /* Not utf8: Inline the first character, for speed. */
4182 if (UCHARAT(s) != nextchr &&
4184 UCHARAT(s) != fold_array[nextchr]))
4186 ln = PL_regoffs[n].end - ln;
4187 if (locinput + ln > PL_regeol)
4189 if (ln > 1 && (type == REF
4190 ? memNE(s, locinput, ln)
4191 : ! folder(s, locinput, ln)))
4194 nextchr = UCHARAT(locinput);
4204 #define ST st->u.eval
4209 regexp_internal *rei;
4210 regnode *startpoint;
4213 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
4214 if (cur_eval && cur_eval->locinput==locinput) {
4215 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
4216 Perl_croak(aTHX_ "Infinite recursion in regex");
4217 if ( ++nochange_depth > max_nochange_depth )
4219 "Pattern subroutine nesting without pos change"
4220 " exceeded limit in regex");
4227 (void)ReREFCNT_inc(rex_sv);
4228 if (OP(scan)==GOSUB) {
4229 startpoint = scan + ARG2L(scan);
4230 ST.close_paren = ARG(scan);
4232 startpoint = rei->program+1;
4235 goto eval_recurse_doit;
4237 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
4238 if (cur_eval && cur_eval->locinput==locinput) {
4239 if ( ++nochange_depth > max_nochange_depth )
4240 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
4245 /* execute the code in the {...} */
4247 SV ** const before = SP;
4248 OP_4tree * const oop = PL_op;
4249 COP * const ocurcop = PL_curcop;
4251 char *saved_regeol = PL_regeol;
4252 struct re_save_state saved_state;
4254 /* To not corrupt the existing regex state while executing the
4255 * eval we would normally put it on the save stack, like with
4256 * save_re_context. However, re-evals have a weird scoping so we
4257 * can't just add ENTER/LEAVE here. With that, things like
4259 * (?{$a=2})(a(?{local$a=$a+1}))*aak*c(?{$b=$a})
4261 * would break, as they expect the localisation to be unwound
4262 * only when the re-engine backtracks through the bit that
4265 * What we do instead is just saving the state in a local c
4268 Copy(&PL_reg_state, &saved_state, 1, struct re_save_state);
4271 PL_op = (OP_4tree*)rexi->data->data[n];
4272 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
4273 " re_eval 0x%"UVxf"\n", PTR2UV(PL_op)) );
4274 /* wrap the call in two SAVECOMPPADs. This ensures that
4275 * when the save stack is eventually unwound, all the
4276 * accumulated SAVEt_CLEARSV's will be processed with
4277 * interspersed SAVEt_COMPPAD's to ensure that lexicals
4278 * are cleared in the right pad */
4280 PAD_SAVE_LOCAL(old_comppad, (PAD*)rexi->data->data[n + 2]);
4281 PL_regoffs[0].end = PL_reg_magic->mg_len = locinput - PL_bostr;
4284 SV *sv_mrk = get_sv("REGMARK", 1);
4285 sv_setsv(sv_mrk, sv_yes_mark);
4288 CALLRUNOPS(aTHX); /* Scalar context. */
4291 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
4297 Copy(&saved_state, &PL_reg_state, 1, struct re_save_state);
4301 PAD_RESTORE_LOCAL(old_comppad);
4302 PL_curcop = ocurcop;
4303 PL_regeol = saved_regeol;
4306 sv_setsv(save_scalar(PL_replgv), ret);
4310 if (logical == 2) { /* Postponed subexpression: /(??{...})/ */
4313 /* extract RE object from returned value; compiling if
4319 SV *const sv = SvRV(ret);
4321 if (SvTYPE(sv) == SVt_REGEXP) {
4323 } else if (SvSMAGICAL(sv)) {
4324 mg = mg_find(sv, PERL_MAGIC_qr);
4327 } else if (SvTYPE(ret) == SVt_REGEXP) {
4329 } else if (SvSMAGICAL(ret)) {
4330 if (SvGMAGICAL(ret)) {
4331 /* I don't believe that there is ever qr magic
4333 assert(!mg_find(ret, PERL_MAGIC_qr));
4334 sv_unmagic(ret, PERL_MAGIC_qr);
4337 mg = mg_find(ret, PERL_MAGIC_qr);
4338 /* testing suggests mg only ends up non-NULL for
4339 scalars who were upgraded and compiled in the
4340 else block below. In turn, this is only
4341 triggered in the "postponed utf8 string" tests
4347 rx = (REGEXP *) mg->mg_obj; /*XXX:dmq*/
4351 rx = reg_temp_copy(NULL, rx);
4355 const I32 osize = PL_regsize;
4358 assert (SvUTF8(ret));
4359 } else if (SvUTF8(ret)) {
4360 /* Not doing UTF-8, despite what the SV says. Is
4361 this only if we're trapped in use 'bytes'? */
4362 /* Make a copy of the octet sequence, but without
4363 the flag on, as the compiler now honours the
4364 SvUTF8 flag on ret. */
4366 const char *const p = SvPV(ret, len);
4367 ret = newSVpvn_flags(p, len, SVs_TEMP);
4369 rx = CALLREGCOMP(ret, pm_flags);
4371 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
4373 /* This isn't a first class regexp. Instead, it's
4374 caching a regexp onto an existing, Perl visible
4376 sv_magic(ret, MUTABLE_SV(rx), PERL_MAGIC_qr, 0, 0);
4381 re = (struct regexp *)SvANY(rx);
4383 RXp_MATCH_COPIED_off(re);
4384 re->subbeg = rex->subbeg;
4385 re->sublen = rex->sublen;
4388 debug_start_match(re_sv, utf8_target, locinput, PL_regeol,
4389 "Matching embedded");
4391 startpoint = rei->program + 1;
4392 ST.close_paren = 0; /* only used for GOSUB */
4393 /* borrowed from regtry */
4394 if (PL_reg_start_tmpl <= re->nparens) {
4395 PL_reg_start_tmpl = re->nparens*3/2 + 3;
4396 if(PL_reg_start_tmp)
4397 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
4399 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
4402 eval_recurse_doit: /* Share code with GOSUB below this line */
4403 /* run the pattern returned from (??{...}) */
4404 ST.cp = regcppush(0); /* Save *all* the positions. */
4405 REGCP_SET(ST.lastcp);
4407 PL_regoffs = re->offs; /* essentially NOOP on GOSUB */
4409 /* see regtry, specifically PL_reglast(?:close)?paren is a pointer! (i dont know why) :dmq */
4410 PL_reglastparen = &re->lastparen;
4411 PL_reglastcloseparen = &re->lastcloseparen;
4413 re->lastcloseparen = 0;
4415 PL_reginput = locinput;
4418 /* XXXX This is too dramatic a measure... */
4421 ST.toggle_reg_flags = PL_reg_flags;
4423 PL_reg_flags |= RF_utf8;
4425 PL_reg_flags &= ~RF_utf8;
4426 ST.toggle_reg_flags ^= PL_reg_flags; /* diff of old and new */
4428 ST.prev_rex = rex_sv;
4429 ST.prev_curlyx = cur_curlyx;
4430 SETREX(rex_sv,re_sv);
4435 ST.prev_eval = cur_eval;
4437 /* now continue from first node in postoned RE */
4438 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint);
4441 /* logical is 1, /(?(?{...})X|Y)/ */
4442 sw = cBOOL(SvTRUE(ret));
4447 case EVAL_AB: /* cleanup after a successful (??{A})B */
4448 /* note: this is called twice; first after popping B, then A */
4449 PL_reg_flags ^= ST.toggle_reg_flags;
4450 ReREFCNT_dec(rex_sv);
4451 SETREX(rex_sv,ST.prev_rex);
4452 rex = (struct regexp *)SvANY(rex_sv);
4453 rexi = RXi_GET(rex);
4455 cur_eval = ST.prev_eval;
4456 cur_curlyx = ST.prev_curlyx;
4458 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
4459 PL_reglastparen = &rex->lastparen;
4460 PL_reglastcloseparen = &rex->lastcloseparen;
4461 /* also update PL_regoffs */
4462 PL_regoffs = rex->offs;
4464 /* XXXX This is too dramatic a measure... */
4466 if ( nochange_depth )
4471 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
4472 /* note: this is called twice; first after popping B, then A */
4473 PL_reg_flags ^= ST.toggle_reg_flags;
4474 ReREFCNT_dec(rex_sv);
4475 SETREX(rex_sv,ST.prev_rex);
4476 rex = (struct regexp *)SvANY(rex_sv);
4477 rexi = RXi_GET(rex);
4478 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
4479 PL_reglastparen = &rex->lastparen;
4480 PL_reglastcloseparen = &rex->lastcloseparen;
4482 PL_reginput = locinput;
4483 REGCP_UNWIND(ST.lastcp);
4485 cur_eval = ST.prev_eval;
4486 cur_curlyx = ST.prev_curlyx;
4487 /* XXXX This is too dramatic a measure... */
4489 if ( nochange_depth )
4495 n = ARG(scan); /* which paren pair */
4496 PL_reg_start_tmp[n] = locinput;
4502 n = ARG(scan); /* which paren pair */
4503 PL_regoffs[n].start = PL_reg_start_tmp[n] - PL_bostr;
4504 PL_regoffs[n].end = locinput - PL_bostr;
4505 /*if (n > PL_regsize)
4507 if (n > *PL_reglastparen)
4508 *PL_reglastparen = n;
4509 *PL_reglastcloseparen = n;
4510 if (cur_eval && cur_eval->u.eval.close_paren == n) {
4518 cursor && OP(cursor)!=END;
4519 cursor=regnext(cursor))
4521 if ( OP(cursor)==CLOSE ){
4523 if ( n <= lastopen ) {
4525 = PL_reg_start_tmp[n] - PL_bostr;
4526 PL_regoffs[n].end = locinput - PL_bostr;
4527 /*if (n > PL_regsize)
4529 if (n > *PL_reglastparen)
4530 *PL_reglastparen = n;
4531 *PL_reglastcloseparen = n;
4532 if ( n == ARG(scan) || (cur_eval &&
4533 cur_eval->u.eval.close_paren == n))
4542 n = ARG(scan); /* which paren pair */
4543 sw = cBOOL(*PL_reglastparen >= n && PL_regoffs[n].end != -1);
4546 /* reg_check_named_buff_matched returns 0 for no match */
4547 sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan));
4551 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
4557 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
4559 next = NEXTOPER(NEXTOPER(scan));
4561 next = scan + ARG(scan);
4562 if (OP(next) == IFTHEN) /* Fake one. */
4563 next = NEXTOPER(NEXTOPER(next));
4567 logical = scan->flags;
4570 /*******************************************************************
4572 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
4573 pattern, where A and B are subpatterns. (For simple A, CURLYM or
4574 STAR/PLUS/CURLY/CURLYN are used instead.)
4576 A*B is compiled as <CURLYX><A><WHILEM><B>
4578 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
4579 state, which contains the current count, initialised to -1. It also sets
4580 cur_curlyx to point to this state, with any previous value saved in the
4583 CURLYX then jumps straight to the WHILEM op, rather than executing A,
4584 since the pattern may possibly match zero times (i.e. it's a while {} loop
4585 rather than a do {} while loop).
4587 Each entry to WHILEM represents a successful match of A. The count in the
4588 CURLYX block is incremented, another WHILEM state is pushed, and execution
4589 passes to A or B depending on greediness and the current count.
4591 For example, if matching against the string a1a2a3b (where the aN are
4592 substrings that match /A/), then the match progresses as follows: (the
4593 pushed states are interspersed with the bits of strings matched so far):
4596 <CURLYX cnt=0><WHILEM>
4597 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
4598 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
4599 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
4600 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
4602 (Contrast this with something like CURLYM, which maintains only a single
4606 a1 <CURLYM cnt=1> a2
4607 a1 a2 <CURLYM cnt=2> a3
4608 a1 a2 a3 <CURLYM cnt=3> b
4611 Each WHILEM state block marks a point to backtrack to upon partial failure
4612 of A or B, and also contains some minor state data related to that
4613 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
4614 overall state, such as the count, and pointers to the A and B ops.
4616 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
4617 must always point to the *current* CURLYX block, the rules are:
4619 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
4620 and set cur_curlyx to point the new block.
4622 When popping the CURLYX block after a successful or unsuccessful match,
4623 restore the previous cur_curlyx.
4625 When WHILEM is about to execute B, save the current cur_curlyx, and set it
4626 to the outer one saved in the CURLYX block.
4628 When popping the WHILEM block after a successful or unsuccessful B match,
4629 restore the previous cur_curlyx.
4631 Here's an example for the pattern (AI* BI)*BO
4632 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
4635 curlyx backtrack stack
4636 ------ ---------------
4638 CO <CO prev=NULL> <WO>
4639 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4640 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4641 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
4643 At this point the pattern succeeds, and we work back down the stack to
4644 clean up, restoring as we go:
4646 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4647 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4648 CO <CO prev=NULL> <WO>
4651 *******************************************************************/
4653 #define ST st->u.curlyx
4655 case CURLYX: /* start of /A*B/ (for complex A) */
4657 /* No need to save/restore up to this paren */
4658 I32 parenfloor = scan->flags;
4660 assert(next); /* keep Coverity happy */
4661 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
4664 /* XXXX Probably it is better to teach regpush to support
4665 parenfloor > PL_regsize... */
4666 if (parenfloor > (I32)*PL_reglastparen)
4667 parenfloor = *PL_reglastparen; /* Pessimization... */
4669 ST.prev_curlyx= cur_curlyx;
4671 ST.cp = PL_savestack_ix;
4673 /* these fields contain the state of the current curly.
4674 * they are accessed by subsequent WHILEMs */
4675 ST.parenfloor = parenfloor;
4680 ST.count = -1; /* this will be updated by WHILEM */
4681 ST.lastloc = NULL; /* this will be updated by WHILEM */
4683 PL_reginput = locinput;
4684 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next));
4688 case CURLYX_end: /* just finished matching all of A*B */
4689 cur_curlyx = ST.prev_curlyx;
4693 case CURLYX_end_fail: /* just failed to match all of A*B */
4695 cur_curlyx = ST.prev_curlyx;
4701 #define ST st->u.whilem
4703 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
4705 /* see the discussion above about CURLYX/WHILEM */
4707 int min = ARG1(cur_curlyx->u.curlyx.me);
4708 int max = ARG2(cur_curlyx->u.curlyx.me);
4709 regnode *A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS;
4711 assert(cur_curlyx); /* keep Coverity happy */
4712 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
4713 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
4714 ST.cache_offset = 0;
4717 PL_reginput = locinput;
4719 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4720 "%*s whilem: matched %ld out of %d..%d\n",
4721 REPORT_CODE_OFF+depth*2, "", (long)n, min, max)
4724 /* First just match a string of min A's. */
4727 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4728 cur_curlyx->u.curlyx.lastloc = locinput;
4729 REGCP_SET(ST.lastcp);
4731 PUSH_STATE_GOTO(WHILEM_A_pre, A);
4735 /* If degenerate A matches "", assume A done. */
4737 if (locinput == cur_curlyx->u.curlyx.lastloc) {
4738 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4739 "%*s whilem: empty match detected, trying continuation...\n",
4740 REPORT_CODE_OFF+depth*2, "")
4742 goto do_whilem_B_max;
4745 /* super-linear cache processing */
4749 if (!PL_reg_maxiter) {
4750 /* start the countdown: Postpone detection until we
4751 * know the match is not *that* much linear. */
4752 PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
4753 /* possible overflow for long strings and many CURLYX's */
4754 if (PL_reg_maxiter < 0)
4755 PL_reg_maxiter = I32_MAX;
4756 PL_reg_leftiter = PL_reg_maxiter;
4759 if (PL_reg_leftiter-- == 0) {
4760 /* initialise cache */
4761 const I32 size = (PL_reg_maxiter + 7)/8;
4762 if (PL_reg_poscache) {
4763 if ((I32)PL_reg_poscache_size < size) {
4764 Renew(PL_reg_poscache, size, char);
4765 PL_reg_poscache_size = size;
4767 Zero(PL_reg_poscache, size, char);
4770 PL_reg_poscache_size = size;
4771 Newxz(PL_reg_poscache, size, char);
4773 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4774 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
4775 PL_colors[4], PL_colors[5])
4779 if (PL_reg_leftiter < 0) {
4780 /* have we already failed at this position? */
4782 offset = (scan->flags & 0xf) - 1
4783 + (locinput - PL_bostr) * (scan->flags>>4);
4784 mask = 1 << (offset % 8);
4786 if (PL_reg_poscache[offset] & mask) {
4787 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4788 "%*s whilem: (cache) already tried at this position...\n",
4789 REPORT_CODE_OFF+depth*2, "")
4791 sayNO; /* cache records failure */
4793 ST.cache_offset = offset;
4794 ST.cache_mask = mask;
4798 /* Prefer B over A for minimal matching. */
4800 if (cur_curlyx->u.curlyx.minmod) {
4801 ST.save_curlyx = cur_curlyx;
4802 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4803 ST.cp = regcppush(ST.save_curlyx->u.curlyx.parenfloor);
4804 REGCP_SET(ST.lastcp);
4805 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B);
4809 /* Prefer A over B for maximal matching. */
4811 if (n < max) { /* More greed allowed? */
4812 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4813 cur_curlyx->u.curlyx.lastloc = locinput;
4814 REGCP_SET(ST.lastcp);
4815 PUSH_STATE_GOTO(WHILEM_A_max, A);
4818 goto do_whilem_B_max;
4822 case WHILEM_B_min: /* just matched B in a minimal match */
4823 case WHILEM_B_max: /* just matched B in a maximal match */
4824 cur_curlyx = ST.save_curlyx;
4828 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
4829 cur_curlyx = ST.save_curlyx;
4830 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4831 cur_curlyx->u.curlyx.count--;
4835 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
4837 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
4838 REGCP_UNWIND(ST.lastcp);
4840 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4841 cur_curlyx->u.curlyx.count--;
4845 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
4846 REGCP_UNWIND(ST.lastcp);
4847 regcppop(rex); /* Restore some previous $<digit>s? */
4848 PL_reginput = locinput;
4849 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4850 "%*s whilem: failed, trying continuation...\n",
4851 REPORT_CODE_OFF+depth*2, "")
4854 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4855 && ckWARN(WARN_REGEXP)
4856 && !(PL_reg_flags & RF_warned))
4858 PL_reg_flags |= RF_warned;
4859 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
4860 "Complex regular subexpression recursion limit (%d) "
4866 ST.save_curlyx = cur_curlyx;
4867 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4868 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B);
4871 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
4872 cur_curlyx = ST.save_curlyx;
4873 REGCP_UNWIND(ST.lastcp);
4876 if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) {
4877 /* Maximum greed exceeded */
4878 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4879 && ckWARN(WARN_REGEXP)
4880 && !(PL_reg_flags & RF_warned))
4882 PL_reg_flags |= RF_warned;
4883 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
4884 "Complex regular subexpression recursion "
4885 "limit (%d) exceeded",
4888 cur_curlyx->u.curlyx.count--;
4892 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4893 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
4895 /* Try grabbing another A and see if it helps. */
4896 PL_reginput = locinput;
4897 cur_curlyx->u.curlyx.lastloc = locinput;
4898 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4899 REGCP_SET(ST.lastcp);
4900 PUSH_STATE_GOTO(WHILEM_A_min,
4901 /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS);
4905 #define ST st->u.branch
4907 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
4908 next = scan + ARG(scan);
4911 scan = NEXTOPER(scan);
4914 case BRANCH: /* /(...|A|...)/ */
4915 scan = NEXTOPER(scan); /* scan now points to inner node */
4916 ST.lastparen = *PL_reglastparen;
4917 ST.next_branch = next;
4919 PL_reginput = locinput;
4921 /* Now go into the branch */
4923 PUSH_YES_STATE_GOTO(BRANCH_next, scan);
4925 PUSH_STATE_GOTO(BRANCH_next, scan);
4929 PL_reginput = locinput;
4930 sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
4931 MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
4932 PUSH_STATE_GOTO(CUTGROUP_next,next);
4934 case CUTGROUP_next_fail:
4937 if (st->u.mark.mark_name)
4938 sv_commit = st->u.mark.mark_name;
4944 case BRANCH_next_fail: /* that branch failed; try the next, if any */
4949 REGCP_UNWIND(ST.cp);
4950 for (n = *PL_reglastparen; n > ST.lastparen; n--)
4951 PL_regoffs[n].end = -1;
4952 *PL_reglastparen = n;
4953 /*dmq: *PL_reglastcloseparen = n; */
4954 scan = ST.next_branch;
4955 /* no more branches? */
4956 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
4958 PerlIO_printf( Perl_debug_log,
4959 "%*s %sBRANCH failed...%s\n",
4960 REPORT_CODE_OFF+depth*2, "",
4966 continue; /* execute next BRANCH[J] op */
4974 #define ST st->u.curlym
4976 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
4978 /* This is an optimisation of CURLYX that enables us to push
4979 * only a single backtracking state, no matter how many matches
4980 * there are in {m,n}. It relies on the pattern being constant
4981 * length, with no parens to influence future backrefs
4985 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
4987 /* if paren positive, emulate an OPEN/CLOSE around A */
4989 U32 paren = ST.me->flags;
4990 if (paren > PL_regsize)
4992 if (paren > *PL_reglastparen)
4993 *PL_reglastparen = paren;
4994 scan += NEXT_OFF(scan); /* Skip former OPEN. */
5002 ST.c1 = CHRTEST_UNINIT;
5005 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
5008 curlym_do_A: /* execute the A in /A{m,n}B/ */
5009 PL_reginput = locinput;
5010 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A); /* match A */
5013 case CURLYM_A: /* we've just matched an A */
5014 locinput = st->locinput;
5015 nextchr = UCHARAT(locinput);
5018 /* after first match, determine A's length: u.curlym.alen */
5019 if (ST.count == 1) {
5020 if (PL_reg_match_utf8) {
5022 while (s < PL_reginput) {
5028 ST.alen = PL_reginput - locinput;
5031 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
5034 PerlIO_printf(Perl_debug_log,
5035 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
5036 (int)(REPORT_CODE_OFF+(depth*2)), "",
5037 (IV) ST.count, (IV)ST.alen)
5040 locinput = PL_reginput;
5042 if (cur_eval && cur_eval->u.eval.close_paren &&
5043 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
5047 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
5048 if ( max == REG_INFTY || ST.count < max )
5049 goto curlym_do_A; /* try to match another A */
5051 goto curlym_do_B; /* try to match B */
5053 case CURLYM_A_fail: /* just failed to match an A */
5054 REGCP_UNWIND(ST.cp);
5056 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
5057 || (cur_eval && cur_eval->u.eval.close_paren &&
5058 cur_eval->u.eval.close_paren == (U32)ST.me->flags))
5061 curlym_do_B: /* execute the B in /A{m,n}B/ */
5062 PL_reginput = locinput;
5063 if (ST.c1 == CHRTEST_UNINIT) {
5064 /* calculate c1 and c2 for possible match of 1st char
5065 * following curly */
5066 ST.c1 = ST.c2 = CHRTEST_VOID;
5067 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
5068 regnode *text_node = ST.B;
5069 if (! HAS_TEXT(text_node))
5070 FIND_NEXT_IMPT(text_node);
5073 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT)
5075 But the former is redundant in light of the latter.
5077 if this changes back then the macro for
5078 IS_TEXT and friends need to change.
5080 if (PL_regkind[OP(text_node)] == EXACT)
5083 ST.c1 = (U8)*STRING(text_node);
5084 switch (OP(text_node)) {
5085 case EXACTF: ST.c2 = PL_fold[ST.c1]; break;
5088 case EXACTFU_NO_TRIE:
5089 case EXACTFU: ST.c2 = PL_fold_latin1[ST.c1]; break;
5090 case EXACTFL: ST.c2 = PL_fold_locale[ST.c1]; break;
5091 default: ST.c2 = ST.c1;
5098 PerlIO_printf(Perl_debug_log,
5099 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
5100 (int)(REPORT_CODE_OFF+(depth*2)),
5103 if (ST.c1 != CHRTEST_VOID
5104 && UCHARAT(PL_reginput) != ST.c1
5105 && UCHARAT(PL_reginput) != ST.c2)
5107 /* simulate B failing */
5109 PerlIO_printf(Perl_debug_log,
5110 "%*s CURLYM Fast bail c1=%"IVdf" c2=%"IVdf"\n",
5111 (int)(REPORT_CODE_OFF+(depth*2)),"",
5114 state_num = CURLYM_B_fail;
5115 goto reenter_switch;
5119 /* mark current A as captured */
5120 I32 paren = ST.me->flags;
5122 PL_regoffs[paren].start
5123 = HOPc(PL_reginput, -ST.alen) - PL_bostr;
5124 PL_regoffs[paren].end = PL_reginput - PL_bostr;
5125 /*dmq: *PL_reglastcloseparen = paren; */
5128 PL_regoffs[paren].end = -1;
5129 if (cur_eval && cur_eval->u.eval.close_paren &&
5130 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
5139 PUSH_STATE_GOTO(CURLYM_B, ST.B); /* match B */
5142 case CURLYM_B_fail: /* just failed to match a B */
5143 REGCP_UNWIND(ST.cp);
5145 I32 max = ARG2(ST.me);
5146 if (max != REG_INFTY && ST.count == max)
5148 goto curlym_do_A; /* try to match a further A */
5150 /* backtrack one A */
5151 if (ST.count == ARG1(ST.me) /* min */)
5154 locinput = HOPc(locinput, -ST.alen);
5155 goto curlym_do_B; /* try to match B */
5158 #define ST st->u.curly
5160 #define CURLY_SETPAREN(paren, success) \
5163 PL_regoffs[paren].start = HOPc(locinput, -1) - PL_bostr; \
5164 PL_regoffs[paren].end = locinput - PL_bostr; \
5165 *PL_reglastcloseparen = paren; \
5168 PL_regoffs[paren].end = -1; \
5171 case STAR: /* /A*B/ where A is width 1 */
5175 scan = NEXTOPER(scan);
5177 case PLUS: /* /A+B/ where A is width 1 */
5181 scan = NEXTOPER(scan);
5183 case CURLYN: /* /(A){m,n}B/ where A is width 1 */
5184 ST.paren = scan->flags; /* Which paren to set */
5185 if (ST.paren > PL_regsize)
5186 PL_regsize = ST.paren;
5187 if (ST.paren > *PL_reglastparen)
5188 *PL_reglastparen = ST.paren;
5189 ST.min = ARG1(scan); /* min to match */
5190 ST.max = ARG2(scan); /* max to match */
5191 if (cur_eval && cur_eval->u.eval.close_paren &&
5192 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5196 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
5198 case CURLY: /* /A{m,n}B/ where A is width 1 */
5200 ST.min = ARG1(scan); /* min to match */
5201 ST.max = ARG2(scan); /* max to match */
5202 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
5205 * Lookahead to avoid useless match attempts
5206 * when we know what character comes next.
5208 * Used to only do .*x and .*?x, but now it allows
5209 * for )'s, ('s and (?{ ... })'s to be in the way
5210 * of the quantifier and the EXACT-like node. -- japhy
5213 if (ST.min > ST.max) /* XXX make this a compile-time check? */
5215 if (HAS_TEXT(next) || JUMPABLE(next)) {
5217 regnode *text_node = next;
5219 if (! HAS_TEXT(text_node))
5220 FIND_NEXT_IMPT(text_node);
5222 if (! HAS_TEXT(text_node))
5223 ST.c1 = ST.c2 = CHRTEST_VOID;
5225 if ( PL_regkind[OP(text_node)] != EXACT ) {
5226 ST.c1 = ST.c2 = CHRTEST_VOID;
5227 goto assume_ok_easy;
5230 s = (U8*)STRING(text_node);
5232 /* Currently we only get here when
5234 PL_rekind[OP(text_node)] == EXACT
5236 if this changes back then the macro for IS_TEXT and
5237 friends need to change. */
5240 switch (OP(text_node)) {
5241 case EXACTF: ST.c2 = PL_fold[ST.c1]; break;
5244 case EXACTFU_NO_TRIE:
5245 case EXACTFU: ST.c2 = PL_fold_latin1[ST.c1]; break;
5246 case EXACTFL: ST.c2 = PL_fold_locale[ST.c1]; break;
5247 default: ST.c2 = ST.c1; break;
5250 else { /* UTF_PATTERN */
5251 if (IS_TEXTFU(text_node) || IS_TEXTF(text_node)) {
5252 STRLEN ulen1, ulen2;
5253 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
5254 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
5256 to_utf8_lower((U8*)s, tmpbuf1, &ulen1);
5257 to_utf8_upper((U8*)s, tmpbuf2, &ulen2);
5259 ST.c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXLEN, 0,
5261 0 : UTF8_ALLOW_ANY);
5262 ST.c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXLEN, 0,
5264 0 : UTF8_ALLOW_ANY);
5266 ST.c1 = utf8n_to_uvuni(tmpbuf1, UTF8_MAXBYTES, 0,
5268 ST.c2 = utf8n_to_uvuni(tmpbuf2, UTF8_MAXBYTES, 0,
5273 ST.c2 = ST.c1 = utf8n_to_uvchr(s, UTF8_MAXBYTES, 0,
5280 ST.c1 = ST.c2 = CHRTEST_VOID;
5285 PL_reginput = locinput;
5288 if (ST.min && regrepeat(rex, ST.A, ST.min, depth) < ST.min)
5291 locinput = PL_reginput;
5293 if (ST.c1 == CHRTEST_VOID)
5294 goto curly_try_B_min;
5296 ST.oldloc = locinput;
5298 /* set ST.maxpos to the furthest point along the
5299 * string that could possibly match */
5300 if (ST.max == REG_INFTY) {
5301 ST.maxpos = PL_regeol - 1;
5303 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
5306 else if (utf8_target) {
5307 int m = ST.max - ST.min;
5308 for (ST.maxpos = locinput;
5309 m >0 && ST.maxpos + UTF8SKIP(ST.maxpos) <= PL_regeol; m--)
5310 ST.maxpos += UTF8SKIP(ST.maxpos);
5313 ST.maxpos = locinput + ST.max - ST.min;
5314 if (ST.maxpos >= PL_regeol)
5315 ST.maxpos = PL_regeol - 1;
5317 goto curly_try_B_min_known;
5321 ST.count = regrepeat(rex, ST.A, ST.max, depth);
5322 locinput = PL_reginput;
5323 if (ST.count < ST.min)
5325 if ((ST.count > ST.min)
5326 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
5328 /* A{m,n} must come at the end of the string, there's
5329 * no point in backing off ... */
5331 /* ...except that $ and \Z can match before *and* after
5332 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
5333 We may back off by one in this case. */
5334 if (UCHARAT(PL_reginput - 1) == '\n' && OP(ST.B) != EOS)
5338 goto curly_try_B_max;
5343 case CURLY_B_min_known_fail:
5344 /* failed to find B in a non-greedy match where c1,c2 valid */
5345 if (ST.paren && ST.count)
5346 PL_regoffs[ST.paren].end = -1;
5348 PL_reginput = locinput; /* Could be reset... */
5349 REGCP_UNWIND(ST.cp);
5350 /* Couldn't or didn't -- move forward. */
5351 ST.oldloc = locinput;
5353 locinput += UTF8SKIP(locinput);
5357 curly_try_B_min_known:
5358 /* find the next place where 'B' could work, then call B */
5362 n = (ST.oldloc == locinput) ? 0 : 1;
5363 if (ST.c1 == ST.c2) {
5365 /* set n to utf8_distance(oldloc, locinput) */
5366 while (locinput <= ST.maxpos &&
5367 utf8n_to_uvchr((U8*)locinput,
5368 UTF8_MAXBYTES, &len,
5369 uniflags) != (UV)ST.c1) {
5375 /* set n to utf8_distance(oldloc, locinput) */
5376 while (locinput <= ST.maxpos) {
5378 const UV c = utf8n_to_uvchr((U8*)locinput,
5379 UTF8_MAXBYTES, &len,
5381 if (c == (UV)ST.c1 || c == (UV)ST.c2)
5389 if (ST.c1 == ST.c2) {
5390 while (locinput <= ST.maxpos &&
5391 UCHARAT(locinput) != ST.c1)
5395 while (locinput <= ST.maxpos
5396 && UCHARAT(locinput) != ST.c1
5397 && UCHARAT(locinput) != ST.c2)
5400 n = locinput - ST.oldloc;
5402 if (locinput > ST.maxpos)
5404 /* PL_reginput == oldloc now */
5407 if (regrepeat(rex, ST.A, n, depth) < n)
5410 PL_reginput = locinput;
5411 CURLY_SETPAREN(ST.paren, ST.count);
5412 if (cur_eval && cur_eval->u.eval.close_paren &&
5413 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5416 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B);
5421 case CURLY_B_min_fail:
5422 /* failed to find B in a non-greedy match where c1,c2 invalid */
5423 if (ST.paren && ST.count)
5424 PL_regoffs[ST.paren].end = -1;
5426 REGCP_UNWIND(ST.cp);
5427 /* failed -- move forward one */
5428 PL_reginput = locinput;
5429 if (regrepeat(rex, ST.A, 1, depth)) {
5431 locinput = PL_reginput;
5432 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
5433 ST.count > 0)) /* count overflow ? */
5436 CURLY_SETPAREN(ST.paren, ST.count);
5437 if (cur_eval && cur_eval->u.eval.close_paren &&
5438 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5441 PUSH_STATE_GOTO(CURLY_B_min, ST.B);
5449 /* a successful greedy match: now try to match B */
5450 if (cur_eval && cur_eval->u.eval.close_paren &&
5451 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5456 if (ST.c1 != CHRTEST_VOID)
5457 c = utf8_target ? utf8n_to_uvchr((U8*)PL_reginput,
5458 UTF8_MAXBYTES, 0, uniflags)
5459 : (UV) UCHARAT(PL_reginput);
5460 /* If it could work, try it. */
5461 if (ST.c1 == CHRTEST_VOID || c == (UV)ST.c1 || c == (UV)ST.c2) {
5462 CURLY_SETPAREN(ST.paren, ST.count);
5463 PUSH_STATE_GOTO(CURLY_B_max, ST.B);
5468 case CURLY_B_max_fail:
5469 /* failed to find B in a greedy match */
5470 if (ST.paren && ST.count)
5471 PL_regoffs[ST.paren].end = -1;
5473 REGCP_UNWIND(ST.cp);
5475 if (--ST.count < ST.min)
5477 PL_reginput = locinput = HOPc(locinput, -1);
5478 goto curly_try_B_max;
5485 /* we've just finished A in /(??{A})B/; now continue with B */
5487 st->u.eval.toggle_reg_flags
5488 = cur_eval->u.eval.toggle_reg_flags;
5489 PL_reg_flags ^= st->u.eval.toggle_reg_flags;
5491 st->u.eval.prev_rex = rex_sv; /* inner */
5492 SETREX(rex_sv,cur_eval->u.eval.prev_rex);
5493 rex = (struct regexp *)SvANY(rex_sv);
5494 rexi = RXi_GET(rex);
5495 cur_curlyx = cur_eval->u.eval.prev_curlyx;
5496 (void)ReREFCNT_inc(rex_sv);
5497 st->u.eval.cp = regcppush(0); /* Save *all* the positions. */
5499 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
5500 PL_reglastparen = &rex->lastparen;
5501 PL_reglastcloseparen = &rex->lastcloseparen;
5503 REGCP_SET(st->u.eval.lastcp);
5504 PL_reginput = locinput;
5506 /* Restore parens of the outer rex without popping the
5508 tmpix = PL_savestack_ix;
5509 PL_savestack_ix = cur_eval->u.eval.lastcp;
5511 PL_savestack_ix = tmpix;
5513 st->u.eval.prev_eval = cur_eval;
5514 cur_eval = cur_eval->u.eval.prev_eval;
5516 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n",
5517 REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval)););
5518 if ( nochange_depth )
5521 PUSH_YES_STATE_GOTO(EVAL_AB,
5522 st->u.eval.prev_eval->u.eval.B); /* match B */
5525 if (locinput < reginfo->till) {
5526 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5527 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
5529 (long)(locinput - PL_reg_starttry),
5530 (long)(reginfo->till - PL_reg_starttry),
5533 sayNO_SILENT; /* Cannot match: too short. */
5535 PL_reginput = locinput; /* put where regtry can find it */
5536 sayYES; /* Success! */
5538 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
5540 PerlIO_printf(Perl_debug_log,
5541 "%*s %ssubpattern success...%s\n",
5542 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
5543 PL_reginput = locinput; /* put where regtry can find it */
5544 sayYES; /* Success! */
5547 #define ST st->u.ifmatch
5549 case SUSPEND: /* (?>A) */
5551 PL_reginput = locinput;
5554 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
5556 goto ifmatch_trivial_fail_test;
5558 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
5560 ifmatch_trivial_fail_test:
5562 char * const s = HOPBACKc(locinput, scan->flags);
5567 sw = 1 - cBOOL(ST.wanted);
5571 next = scan + ARG(scan);
5579 PL_reginput = locinput;
5583 ST.logical = logical;
5584 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
5586 /* execute body of (?...A) */
5587 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)));
5590 case IFMATCH_A_fail: /* body of (?...A) failed */
5591 ST.wanted = !ST.wanted;
5594 case IFMATCH_A: /* body of (?...A) succeeded */
5596 sw = cBOOL(ST.wanted);
5598 else if (!ST.wanted)
5601 if (OP(ST.me) == SUSPEND)
5602 locinput = PL_reginput;
5604 locinput = PL_reginput = st->locinput;
5605 nextchr = UCHARAT(locinput);
5607 scan = ST.me + ARG(ST.me);
5610 continue; /* execute B */
5615 next = scan + ARG(scan);
5620 reginfo->cutpoint = PL_regeol;
5623 PL_reginput = locinput;
5625 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5626 PUSH_STATE_GOTO(COMMIT_next,next);
5628 case COMMIT_next_fail:
5635 #define ST st->u.mark
5637 ST.prev_mark = mark_state;
5638 ST.mark_name = sv_commit = sv_yes_mark
5639 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5641 ST.mark_loc = PL_reginput = locinput;
5642 PUSH_YES_STATE_GOTO(MARKPOINT_next,next);
5644 case MARKPOINT_next:
5645 mark_state = ST.prev_mark;
5648 case MARKPOINT_next_fail:
5649 if (popmark && sv_eq(ST.mark_name,popmark))
5651 if (ST.mark_loc > startpoint)
5652 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5653 popmark = NULL; /* we found our mark */
5654 sv_commit = ST.mark_name;
5657 PerlIO_printf(Perl_debug_log,
5658 "%*s %ssetting cutpoint to mark:%"SVf"...%s\n",
5659 REPORT_CODE_OFF+depth*2, "",
5660 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
5663 mark_state = ST.prev_mark;
5664 sv_yes_mark = mark_state ?
5665 mark_state->u.mark.mark_name : NULL;
5669 PL_reginput = locinput;
5671 /* (*SKIP) : if we fail we cut here*/
5672 ST.mark_name = NULL;
5673 ST.mark_loc = locinput;
5674 PUSH_STATE_GOTO(SKIP_next,next);
5676 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
5677 otherwise do nothing. Meaning we need to scan
5679 regmatch_state *cur = mark_state;
5680 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5683 if ( sv_eq( cur->u.mark.mark_name,
5686 ST.mark_name = find;
5687 PUSH_STATE_GOTO( SKIP_next, next );
5689 cur = cur->u.mark.prev_mark;
5692 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
5694 case SKIP_next_fail:
5696 /* (*CUT:NAME) - Set up to search for the name as we
5697 collapse the stack*/
5698 popmark = ST.mark_name;
5700 /* (*CUT) - No name, we cut here.*/
5701 if (ST.mark_loc > startpoint)
5702 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5703 /* but we set sv_commit to latest mark_name if there
5704 is one so they can test to see how things lead to this
5707 sv_commit=mark_state->u.mark.mark_name;
5714 if ((n=is_LNBREAK(locinput,utf8_target))) {
5716 nextchr = UCHARAT(locinput);
5721 #define CASE_CLASS(nAmE) \
5723 if (locinput >= PL_regeol) \
5725 if ((n=is_##nAmE(locinput,utf8_target))) { \
5727 nextchr = UCHARAT(locinput); \
5732 if (locinput >= PL_regeol) \
5734 if ((n=is_##nAmE(locinput,utf8_target))) { \
5737 locinput += UTF8SKIP(locinput); \
5738 nextchr = UCHARAT(locinput); \
5743 CASE_CLASS(HORIZWS);
5747 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
5748 PTR2UV(scan), OP(scan));
5749 Perl_croak(aTHX_ "regexp memory corruption");
5753 /* switch break jumps here */
5754 scan = next; /* prepare to execute the next op and ... */
5755 continue; /* ... jump back to the top, reusing st */
5759 /* push a state that backtracks on success */
5760 st->u.yes.prev_yes_state = yes_state;
5764 /* push a new regex state, then continue at scan */
5766 regmatch_state *newst;
5769 regmatch_state *cur = st;
5770 regmatch_state *curyes = yes_state;
5772 regmatch_slab *slab = PL_regmatch_slab;
5773 for (;curd > -1;cur--,curd--) {
5774 if (cur < SLAB_FIRST(slab)) {
5776 cur = SLAB_LAST(slab);
5778 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n",
5779 REPORT_CODE_OFF + 2 + depth * 2,"",
5780 curd, PL_reg_name[cur->resume_state],
5781 (curyes == cur) ? "yes" : ""
5784 curyes = cur->u.yes.prev_yes_state;
5787 DEBUG_STATE_pp("push")
5790 st->locinput = locinput;
5792 if (newst > SLAB_LAST(PL_regmatch_slab))
5793 newst = S_push_slab(aTHX);
5794 PL_regmatch_state = newst;
5796 locinput = PL_reginput;
5797 nextchr = UCHARAT(locinput);
5805 * We get here only if there's trouble -- normally "case END" is
5806 * the terminating point.
5808 Perl_croak(aTHX_ "corrupted regexp pointers");
5814 /* we have successfully completed a subexpression, but we must now
5815 * pop to the state marked by yes_state and continue from there */
5816 assert(st != yes_state);
5818 while (st != yes_state) {
5820 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5821 PL_regmatch_slab = PL_regmatch_slab->prev;
5822 st = SLAB_LAST(PL_regmatch_slab);
5826 DEBUG_STATE_pp("pop (no final)");
5828 DEBUG_STATE_pp("pop (yes)");
5834 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
5835 || yes_state > SLAB_LAST(PL_regmatch_slab))
5837 /* not in this slab, pop slab */
5838 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
5839 PL_regmatch_slab = PL_regmatch_slab->prev;
5840 st = SLAB_LAST(PL_regmatch_slab);
5842 depth -= (st - yes_state);
5845 yes_state = st->u.yes.prev_yes_state;
5846 PL_regmatch_state = st;
5849 locinput= st->locinput;
5850 nextchr = UCHARAT(locinput);
5852 state_num = st->resume_state + no_final;
5853 goto reenter_switch;
5856 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
5857 PL_colors[4], PL_colors[5]));
5859 if (PL_reg_eval_set) {
5860 /* each successfully executed (?{...}) block does the equivalent of
5861 * local $^R = do {...}
5862 * When popping the save stack, all these locals would be undone;
5863 * bypass this by setting the outermost saved $^R to the latest
5865 if (oreplsv != GvSV(PL_replgv))
5866 sv_setsv(oreplsv, GvSV(PL_replgv));
5873 PerlIO_printf(Perl_debug_log,
5874 "%*s %sfailed...%s\n",
5875 REPORT_CODE_OFF+depth*2, "",
5876 PL_colors[4], PL_colors[5])
5888 /* there's a previous state to backtrack to */
5890 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5891 PL_regmatch_slab = PL_regmatch_slab->prev;
5892 st = SLAB_LAST(PL_regmatch_slab);
5894 PL_regmatch_state = st;
5895 locinput= st->locinput;
5896 nextchr = UCHARAT(locinput);
5898 DEBUG_STATE_pp("pop");
5900 if (yes_state == st)
5901 yes_state = st->u.yes.prev_yes_state;
5903 state_num = st->resume_state + 1; /* failure = success + 1 */
5904 goto reenter_switch;
5909 if (rex->intflags & PREGf_VERBARG_SEEN) {
5910 SV *sv_err = get_sv("REGERROR", 1);
5911 SV *sv_mrk = get_sv("REGMARK", 1);
5913 sv_commit = &PL_sv_no;
5915 sv_yes_mark = &PL_sv_yes;
5918 sv_commit = &PL_sv_yes;
5919 sv_yes_mark = &PL_sv_no;
5921 sv_setsv(sv_err, sv_commit);
5922 sv_setsv(sv_mrk, sv_yes_mark);
5925 /* clean up; in particular, free all slabs above current one */
5926 LEAVE_SCOPE(oldsave);
5932 - regrepeat - repeatedly match something simple, report how many
5935 * [This routine now assumes that it will only match on things of length 1.
5936 * That was true before, but now we assume scan - reginput is the count,
5937 * rather than incrementing count on every character. [Er, except utf8.]]
5940 S_regrepeat(pTHX_ const regexp *prog, const regnode *p, I32 max, int depth)
5943 register char *scan;
5945 register char *loceol = PL_regeol;
5946 register I32 hardcount = 0;
5947 register bool utf8_target = PL_reg_match_utf8;
5950 PERL_UNUSED_ARG(depth);
5953 PERL_ARGS_ASSERT_REGREPEAT;
5956 if (max == REG_INFTY)
5958 else if (max < loceol - scan)
5959 loceol = scan + max;
5964 while (scan < loceol && hardcount < max && *scan != '\n') {
5965 scan += UTF8SKIP(scan);
5969 while (scan < loceol && *scan != '\n')
5976 while (scan < loceol && hardcount < max) {
5977 scan += UTF8SKIP(scan);
5988 /* To get here, EXACTish nodes must have *byte* length == 1. That
5989 * means they match only characters in the string that can be expressed
5990 * as a single byte. For non-utf8 strings, that means a simple match.
5991 * For utf8 strings, the character matched must be an invariant, or
5992 * downgradable to a single byte. The pattern's utf8ness is
5993 * irrelevant, as since it's a single byte, it either isn't utf8, or if
5994 * it is, it's an invariant */
5997 assert(! UTF_PATTERN || UNI_IS_INVARIANT(c));
5999 if (! utf8_target || UNI_IS_INVARIANT(c)) {
6000 while (scan < loceol && UCHARAT(scan) == c) {
6006 /* Here, the string is utf8, and the pattern char is different
6007 * in utf8 than not, so can't compare them directly. Outside the
6008 * loop, find the two utf8 bytes that represent c, and then
6009 * look for those in sequence in the utf8 string */
6010 U8 high = UTF8_TWO_BYTE_HI(c);
6011 U8 low = UTF8_TWO_BYTE_LO(c);
6014 while (hardcount < max
6015 && scan + 1 < loceol
6016 && UCHARAT(scan) == high
6017 && UCHARAT(scan + 1) == low)
6025 utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
6029 PL_reg_flags |= RF_tainted;
6030 utf8_flags = FOLDEQ_UTF8_LOCALE;
6038 case EXACTFU_NO_TRIE:
6040 utf8_flags = (UTF_PATTERN) ? FOLDEQ_S2_ALREADY_FOLDED : 0;
6042 /* The comments for the EXACT case above apply as well to these fold
6047 assert(! UTF_PATTERN || UNI_IS_INVARIANT(c));
6049 if (utf8_target || OP(p) == EXACTFU_SS) { /* Use full Unicode fold matching */
6050 char *tmpeol = loceol;
6051 while (hardcount < max
6052 && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target,
6053 STRING(p), NULL, 1, cBOOL(UTF_PATTERN), utf8_flags))
6060 /* XXX Note that the above handles properly the German sharp s in
6061 * the pattern matching ss in the string. But it doesn't handle
6062 * properly cases where the string contains say 'LIGATURE ff' and
6063 * the pattern is 'f+'. This would require, say, a new function or
6064 * revised interface to foldEQ_utf8(), in which the maximum number
6065 * of characters to match could be passed and it would return how
6066 * many actually did. This is just one of many cases where
6067 * multi-char folds don't work properly, and so the fix is being
6073 /* Here, the string isn't utf8 and c is a single byte; and either
6074 * the pattern isn't utf8 or c is an invariant, so its utf8ness
6075 * doesn't affect c. Can just do simple comparisons for exact or
6078 case EXACTF: folded = PL_fold[c]; break;
6080 case EXACTFU_NO_TRIE:
6081 case EXACTFU: folded = PL_fold_latin1[c]; break;
6082 case EXACTFL: folded = PL_fold_locale[c]; break;
6083 default: Perl_croak(aTHX_ "panic: Unexpected op %u", OP(p));
6085 while (scan < loceol &&
6086 (UCHARAT(scan) == c || UCHARAT(scan) == folded))
6094 if (utf8_target || OP(p) == ANYOFV) {
6097 inclasslen = loceol - scan;
6098 while (hardcount < max
6099 && ((inclasslen = loceol - scan) > 0)
6100 && reginclass(prog, p, (U8*)scan, &inclasslen, utf8_target))
6106 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
6114 LOAD_UTF8_CHARCLASS_ALNUM();
6115 while (hardcount < max && scan < loceol &&
6116 swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target))
6118 scan += UTF8SKIP(scan);
6122 while (scan < loceol && isWORDCHAR_L1((U8) *scan)) {
6130 while (scan < loceol && isALNUM((U8) *scan)) {
6135 while (scan < loceol && isWORDCHAR_A((U8) *scan)) {
6140 PL_reg_flags |= RF_tainted;
6143 while (hardcount < max && scan < loceol &&
6144 isALNUM_LC_utf8((U8*)scan)) {
6145 scan += UTF8SKIP(scan);
6149 while (scan < loceol && isALNUM_LC(*scan))
6159 LOAD_UTF8_CHARCLASS_ALNUM();
6160 while (hardcount < max && scan < loceol &&
6161 ! swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target))
6163 scan += UTF8SKIP(scan);
6167 while (scan < loceol && ! isWORDCHAR_L1((U8) *scan)) {
6174 goto utf8_Nwordchar;
6175 while (scan < loceol && ! isALNUM((U8) *scan)) {
6181 while (scan < loceol && ! isWORDCHAR_A((U8) *scan)) {
6182 scan += UTF8SKIP(scan);
6186 while (scan < loceol && ! isWORDCHAR_A((U8) *scan)) {
6192 PL_reg_flags |= RF_tainted;
6195 while (hardcount < max && scan < loceol &&
6196 !isALNUM_LC_utf8((U8*)scan)) {
6197 scan += UTF8SKIP(scan);
6201 while (scan < loceol && !isALNUM_LC(*scan))
6211 LOAD_UTF8_CHARCLASS_SPACE();
6212 while (hardcount < max && scan < loceol &&
6214 swash_fetch(PL_utf8_space,(U8*)scan, utf8_target)))
6216 scan += UTF8SKIP(scan);
6222 while (scan < loceol && isSPACE_L1((U8) *scan)) {
6231 while (scan < loceol && isSPACE((U8) *scan)) {
6236 while (scan < loceol && isSPACE_A((U8) *scan)) {
6241 PL_reg_flags |= RF_tainted;
6244 while (hardcount < max && scan < loceol &&
6245 isSPACE_LC_utf8((U8*)scan)) {
6246 scan += UTF8SKIP(scan);
6250 while (scan < loceol && isSPACE_LC(*scan))
6260 LOAD_UTF8_CHARCLASS_SPACE();
6261 while (hardcount < max && scan < loceol &&
6263 swash_fetch(PL_utf8_space,(U8*)scan, utf8_target)))
6265 scan += UTF8SKIP(scan);
6271 while (scan < loceol && ! isSPACE_L1((U8) *scan)) {
6280 while (scan < loceol && ! isSPACE((U8) *scan)) {
6286 while (scan < loceol && ! isSPACE_A((U8) *scan)) {
6287 scan += UTF8SKIP(scan);
6291 while (scan < loceol && ! isSPACE_A((U8) *scan)) {
6297 PL_reg_flags |= RF_tainted;
6300 while (hardcount < max && scan < loceol &&
6301 !isSPACE_LC_utf8((U8*)scan)) {
6302 scan += UTF8SKIP(scan);
6306 while (scan < loceol && !isSPACE_LC(*scan))
6313 LOAD_UTF8_CHARCLASS_DIGIT();
6314 while (hardcount < max && scan < loceol &&
6315 swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) {
6316 scan += UTF8SKIP(scan);
6320 while (scan < loceol && isDIGIT(*scan))
6325 while (scan < loceol && isDIGIT_A((U8) *scan)) {
6330 PL_reg_flags |= RF_tainted;
6333 while (hardcount < max && scan < loceol &&
6334 isDIGIT_LC_utf8((U8*)scan)) {
6335 scan += UTF8SKIP(scan);
6339 while (scan < loceol && isDIGIT_LC(*scan))
6346 LOAD_UTF8_CHARCLASS_DIGIT();
6347 while (hardcount < max && scan < loceol &&
6348 !swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) {
6349 scan += UTF8SKIP(scan);
6353 while (scan < loceol && !isDIGIT(*scan))
6359 while (scan < loceol && ! isDIGIT_A((U8) *scan)) {
6360 scan += UTF8SKIP(scan);
6364 while (scan < loceol && ! isDIGIT_A((U8) *scan)) {
6370 PL_reg_flags |= RF_tainted;
6373 while (hardcount < max && scan < loceol &&
6374 !isDIGIT_LC_utf8((U8*)scan)) {
6375 scan += UTF8SKIP(scan);
6379 while (scan < loceol && !isDIGIT_LC(*scan))
6386 while (hardcount < max && scan < loceol && (c=is_LNBREAK_utf8(scan))) {
6392 LNBREAK can match two latin chars, which is ok,
6393 because we have a null terminated string, but we
6394 have to use hardcount in this situation
6396 while (scan < loceol && (c=is_LNBREAK_latin1(scan))) {
6405 while (hardcount < max && scan < loceol && (c=is_HORIZWS_utf8(scan))) {
6410 while (scan < loceol && is_HORIZWS_latin1(scan))
6417 while (hardcount < max && scan < loceol && !is_HORIZWS_utf8(scan)) {
6418 scan += UTF8SKIP(scan);
6422 while (scan < loceol && !is_HORIZWS_latin1(scan))
6430 while (hardcount < max && scan < loceol && (c=is_VERTWS_utf8(scan))) {
6435 while (scan < loceol && is_VERTWS_latin1(scan))
6443 while (hardcount < max && scan < loceol && !is_VERTWS_utf8(scan)) {
6444 scan += UTF8SKIP(scan);
6448 while (scan < loceol && !is_VERTWS_latin1(scan))
6454 default: /* Called on something of 0 width. */
6455 break; /* So match right here or not at all. */
6461 c = scan - PL_reginput;
6465 GET_RE_DEBUG_FLAGS_DECL;
6467 SV * const prop = sv_newmortal();
6468 regprop(prog, prop, p);
6469 PerlIO_printf(Perl_debug_log,
6470 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
6471 REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max);
6479 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
6481 - regclass_swash - prepare the utf8 swash. Wraps the shared core version to
6482 create a copy so that changes the caller makes won't change the shared one
6485 Perl_regclass_swash(pTHX_ const regexp *prog, register const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
6487 PERL_ARGS_ASSERT_REGCLASS_SWASH;
6488 return newSVsv(core_regclass_swash(prog, node, doinit, listsvp, altsvp));
6493 S_core_regclass_swash(pTHX_ const regexp *prog, register const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
6495 /* Returns the swash for the input 'node' in the regex 'prog'.
6496 * If <doinit> is true, will attempt to create the swash if not already
6498 * If <listsvp> is non-null, will return the swash initialization string in
6500 * If <altsvp> is non-null, will return the alternates to the regular swash
6502 * Tied intimately to how regcomp.c sets up the data structure */
6510 RXi_GET_DECL(prog,progi);
6511 const struct reg_data * const data = prog ? progi->data : NULL;
6513 PERL_ARGS_ASSERT_CORE_REGCLASS_SWASH;
6515 assert(ANYOF_NONBITMAP(node));
6517 if (data && data->count) {
6518 const U32 n = ARG(node);
6520 if (data->what[n] == 's') {
6521 SV * const rv = MUTABLE_SV(data->data[n]);
6522 AV * const av = MUTABLE_AV(SvRV(rv));
6523 SV **const ary = AvARRAY(av);
6524 bool invlist_has_user_defined_property;
6526 si = *ary; /* ary[0] = the string to initialize the swash with */
6528 /* Elements 3 and 4 are either both present or both absent. [3] is
6529 * any inversion list generated at compile time; [4] indicates if
6530 * that inversion list has any user-defined properties in it. */
6531 if (av_len(av) >= 3) {
6533 invlist_has_user_defined_property = cBOOL(SvUV(ary[4]));
6537 invlist_has_user_defined_property = FALSE;
6540 /* Element [1] is reserved for the set-up swash. If already there,
6541 * return it; if not, create it and store it there */
6542 if (SvROK(ary[1])) {
6545 else if (si && doinit) {
6547 sw = _core_swash_init("utf8", /* the utf8 package */
6551 0, /* not from tr/// */
6552 FALSE, /* is error if can't find
6555 invlist_has_user_defined_property);
6556 (void)av_store(av, 1, sw);
6559 /* Element [2] is for any multi-char folds. Note that is a
6560 * fundamentally flawed design, because can't backtrack and try
6561 * again. See [perl #89774] */
6562 if (SvTYPE(ary[2]) == SVt_PVAV) {
6569 SV* matches_string = newSVpvn("", 0);
6572 /* Use the swash, if any, which has to have incorporated into it all
6576 && SvTYPE(SvRV(sw)) == SVt_PVHV
6577 && (invlistsvp = hv_fetchs(MUTABLE_HV(SvRV(sw)), "INVLIST", FALSE)))
6579 invlist = *invlistsvp;
6581 else if (si && si != &PL_sv_undef) {
6583 /* If no swash, use the input nitialization string, if available */
6584 sv_catsv(matches_string, si);
6587 /* Add the inversion list to whatever we have. This may have come from
6588 * the swash, or from an input parameter */
6590 sv_catsv(matches_string, _invlist_contents(invlist));
6592 *listsvp = matches_string;
6602 - reginclass - determine if a character falls into a character class
6604 n is the ANYOF regnode
6605 p is the target string
6606 lenp is pointer to the maximum number of bytes of how far to go in p
6607 (This is assumed wthout checking to always be at least the current
6609 utf8_target tells whether p is in UTF-8.
6611 Returns true if matched; false otherwise. If lenp is not NULL, on return
6612 from a successful match, the value it points to will be updated to how many
6613 bytes in p were matched. If there was no match, the value is undefined,
6614 possibly changed from the input.
6616 Note that this can be a synthetic start class, a combination of various
6617 nodes, so things you think might be mutually exclusive, such as locale,
6618 aren't. It can match both locale and non-locale
6623 S_reginclass(pTHX_ const regexp * const prog, register const regnode * const n, register const U8* const p, STRLEN* lenp, register const bool utf8_target)
6626 const char flags = ANYOF_FLAGS(n);
6632 PERL_ARGS_ASSERT_REGINCLASS;
6634 /* If c is not already the code point, get it */
6635 if (utf8_target && !UTF8_IS_INVARIANT(c)) {
6636 c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &c_len,
6637 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV)
6638 | UTF8_ALLOW_FFFF | UTF8_CHECK_ONLY);
6639 /* see [perl #37836] for UTF8_ALLOW_ANYUV; [perl #38293] for
6640 * UTF8_ALLOW_FFFF */
6641 if (c_len == (STRLEN)-1)
6642 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
6648 /* Use passed in max length, or one character if none passed in or less
6649 * than one character. And assume will match just one character. This is
6650 * overwritten later if matched more. */
6652 maxlen = (*lenp > c_len) ? *lenp : c_len;
6660 /* If this character is potentially in the bitmap, check it */
6662 if (ANYOF_BITMAP_TEST(n, c))
6664 else if (flags & ANYOF_NON_UTF8_LATIN1_ALL
6671 else if (flags & ANYOF_LOCALE) {
6672 PL_reg_flags |= RF_tainted;
6674 if ((flags & ANYOF_LOC_NONBITMAP_FOLD)
6675 && ANYOF_BITMAP_TEST(n, PL_fold_locale[c]))
6679 else if (ANYOF_CLASS_TEST_ANY_SET(n) &&
6680 ((ANYOF_CLASS_TEST(n, ANYOF_ALNUM) && isALNUM_LC(c)) ||
6681 (ANYOF_CLASS_TEST(n, ANYOF_NALNUM) && !isALNUM_LC(c)) ||
6682 (ANYOF_CLASS_TEST(n, ANYOF_SPACE) && isSPACE_LC(c)) ||
6683 (ANYOF_CLASS_TEST(n, ANYOF_NSPACE) && !isSPACE_LC(c)) ||
6684 (ANYOF_CLASS_TEST(n, ANYOF_DIGIT) && isDIGIT_LC(c)) ||
6685 (ANYOF_CLASS_TEST(n, ANYOF_NDIGIT) && !isDIGIT_LC(c)) ||
6686 (ANYOF_CLASS_TEST(n, ANYOF_ALNUMC) && isALNUMC_LC(c)) ||
6687 (ANYOF_CLASS_TEST(n, ANYOF_NALNUMC) && !isALNUMC_LC(c)) ||
6688 (ANYOF_CLASS_TEST(n, ANYOF_ALPHA) && isALPHA_LC(c)) ||
6689 (ANYOF_CLASS_TEST(n, ANYOF_NALPHA) && !isALPHA_LC(c)) ||
6690 (ANYOF_CLASS_TEST(n, ANYOF_ASCII) && isASCII_LC(c)) ||
6691 (ANYOF_CLASS_TEST(n, ANYOF_NASCII) && !isASCII_LC(c)) ||
6692 (ANYOF_CLASS_TEST(n, ANYOF_CNTRL) && isCNTRL_LC(c)) ||
6693 (ANYOF_CLASS_TEST(n, ANYOF_NCNTRL) && !isCNTRL_LC(c)) ||
6694 (ANYOF_CLASS_TEST(n, ANYOF_GRAPH) && isGRAPH_LC(c)) ||
6695 (ANYOF_CLASS_TEST(n, ANYOF_NGRAPH) && !isGRAPH_LC(c)) ||
6696 (ANYOF_CLASS_TEST(n, ANYOF_LOWER) && isLOWER_LC(c)) ||
6697 (ANYOF_CLASS_TEST(n, ANYOF_NLOWER) && !isLOWER_LC(c)) ||
6698 (ANYOF_CLASS_TEST(n, ANYOF_PRINT) && isPRINT_LC(c)) ||
6699 (ANYOF_CLASS_TEST(n, ANYOF_NPRINT) && !isPRINT_LC(c)) ||
6700 (ANYOF_CLASS_TEST(n, ANYOF_PUNCT) && isPUNCT_LC(c)) ||
6701 (ANYOF_CLASS_TEST(n, ANYOF_NPUNCT) && !isPUNCT_LC(c)) ||
6702 (ANYOF_CLASS_TEST(n, ANYOF_UPPER) && isUPPER_LC(c)) ||
6703 (ANYOF_CLASS_TEST(n, ANYOF_NUPPER) && !isUPPER_LC(c)) ||
6704 (ANYOF_CLASS_TEST(n, ANYOF_XDIGIT) && isXDIGIT(c)) ||
6705 (ANYOF_CLASS_TEST(n, ANYOF_NXDIGIT) && !isXDIGIT(c)) ||
6706 (ANYOF_CLASS_TEST(n, ANYOF_PSXSPC) && isPSXSPC(c)) ||
6707 (ANYOF_CLASS_TEST(n, ANYOF_NPSXSPC) && !isPSXSPC(c)) ||
6708 (ANYOF_CLASS_TEST(n, ANYOF_BLANK) && isBLANK_LC(c)) ||
6709 (ANYOF_CLASS_TEST(n, ANYOF_NBLANK) && !isBLANK_LC(c))
6710 ) /* How's that for a conditional? */
6717 /* If the bitmap didn't (or couldn't) match, and something outside the
6718 * bitmap could match, try that. Locale nodes specifiy completely the
6719 * behavior of code points in the bit map (otherwise, a utf8 target would
6720 * cause them to be treated as Unicode and not locale), except in
6721 * the very unlikely event when this node is a synthetic start class, which
6722 * could be a combination of locale and non-locale nodes. So allow locale
6723 * to match for the synthetic start class, which will give a false
6724 * positive that will be resolved when the match is done again as not part
6725 * of the synthetic start class */
6727 if (utf8_target && (flags & ANYOF_UNICODE_ALL) && c >= 256) {
6728 match = TRUE; /* Everything above 255 matches */
6730 else if (ANYOF_NONBITMAP(n)
6731 && ((flags & ANYOF_NONBITMAP_NON_UTF8)
6734 || (! (flags & ANYOF_LOCALE))
6735 || (flags & ANYOF_IS_SYNTHETIC)))))
6738 SV * const sw = core_regclass_swash(prog, n, TRUE, 0, (SV**)&av);
6746 /* Not utf8. Convert as much of the string as available up
6747 * to the limit of how far the (single) character in the
6748 * pattern can possibly match (no need to go further). If
6749 * the node is a straight ANYOF or not folding, it can't
6750 * match more than one. Otherwise, It can match up to how
6751 * far a single char can fold to. Since not utf8, each
6752 * character is a single byte, so the max it can be in
6753 * bytes is the same as the max it can be in characters */
6754 STRLEN len = (OP(n) == ANYOF
6755 || ! (flags & ANYOF_LOC_NONBITMAP_FOLD))
6757 : (maxlen < UTF8_MAX_FOLD_CHAR_EXPAND)
6759 : UTF8_MAX_FOLD_CHAR_EXPAND;
6760 utf8_p = bytes_to_utf8(p, &len);
6763 if (swash_fetch(sw, utf8_p, TRUE))
6765 else if (flags & ANYOF_LOC_NONBITMAP_FOLD) {
6767 /* Here, we need to test if the fold of the target string
6768 * matches. The non-multi char folds have all been moved to
6769 * the compilation phase, and the multi-char folds have
6770 * been stored by regcomp into 'av'; we linearly check to
6771 * see if any match the target string (folded). We know
6772 * that the originals were each one character, but we don't
6773 * currently know how many characters/bytes each folded to,
6774 * except we do know that there are small limits imposed by
6775 * Unicode. XXX A performance enhancement would be to have
6776 * regcomp.c store the max number of chars/bytes that are
6777 * in an av entry, as, say the 0th element. Even better
6778 * would be to have a hash of the few characters that can
6779 * start a multi-char fold to the max number of chars of
6782 * If there is a match, we will need to advance (if lenp is
6783 * specified) the match pointer in the target string. But
6784 * what we are comparing here isn't that string directly,
6785 * but its fold, whose length may differ from the original.
6786 * As we go along in constructing the fold, therefore, we
6787 * create a map so that we know how many bytes in the
6788 * source to advance given that we have matched a certain
6789 * number of bytes in the fold. This map is stored in
6790 * 'map_fold_len_back'. Let n mean the number of bytes in
6791 * the fold of the first character that we are folding.
6792 * Then map_fold_len_back[n] is set to the number of bytes
6793 * in that first character. Similarly let m be the
6794 * corresponding number for the second character to be
6795 * folded. Then map_fold_len_back[n+m] is set to the
6796 * number of bytes occupied by the first two source
6797 * characters. ... */
6798 U8 map_fold_len_back[UTF8_MAXBYTES_CASE+1] = { 0 };
6799 U8 folded[UTF8_MAXBYTES_CASE+1];
6800 STRLEN foldlen = 0; /* num bytes in fold of 1st char */
6801 STRLEN total_foldlen = 0; /* num bytes in fold of all
6804 if (OP(n) == ANYOF || maxlen == 1 || ! lenp || ! av) {
6806 /* Here, only need to fold the first char of the target
6807 * string. It the source wasn't utf8, is 1 byte long */
6808 to_utf8_fold(utf8_p, folded, &foldlen);
6809 total_foldlen = foldlen;
6810 map_fold_len_back[foldlen] = (utf8_target)
6816 /* Here, need to fold more than the first char. Do so
6817 * up to the limits */
6818 U8* source_ptr = utf8_p; /* The source for the fold
6821 U8* folded_ptr = folded;
6822 U8* e = utf8_p + maxlen; /* Can't go beyond last
6823 available byte in the
6827 i < UTF8_MAX_FOLD_CHAR_EXPAND && source_ptr < e;
6831 /* Fold the next character */
6832 U8 this_char_folded[UTF8_MAXBYTES_CASE+1];
6833 STRLEN this_char_foldlen;
6834 to_utf8_fold(source_ptr,
6836 &this_char_foldlen);
6838 /* Bail if it would exceed the byte limit for
6839 * folding a single char. */
6840 if (this_char_foldlen + folded_ptr - folded >
6846 /* Add the fold of this character */
6847 Copy(this_char_folded,
6851 source_ptr += UTF8SKIP(source_ptr);
6852 folded_ptr += this_char_foldlen;
6853 total_foldlen = folded_ptr - folded;
6855 /* Create map from the number of bytes in the fold
6856 * back to the number of bytes in the source. If
6857 * the source isn't utf8, the byte count is just
6858 * the number of characters so far */
6859 map_fold_len_back[total_foldlen]
6861 ? source_ptr - utf8_p
6868 /* Do the linear search to see if the fold is in the list
6869 * of multi-char folds. */
6872 for (i = 0; i <= av_len(av); i++) {
6873 SV* const sv = *av_fetch(av, i, FALSE);
6875 const char * const s = SvPV_const(sv, len);
6877 if (len <= total_foldlen
6878 && memEQ(s, (char*)folded, len)
6880 /* If 0, means matched a partial char. See
6882 && map_fold_len_back[len])
6885 /* Advance the target string ptr to account for
6886 * this fold, but have to translate from the
6887 * folded length to the corresponding source
6890 *lenp = map_fold_len_back[len];
6899 /* If we allocated a string above, free it */
6900 if (! utf8_target) Safefree(utf8_p);
6905 return (flags & ANYOF_INVERT) ? !match : match;
6909 S_reghop3(U8 *s, I32 off, const U8* lim)
6911 /* return the position 'off' UTF-8 characters away from 's', forward if
6912 * 'off' >= 0, backwards if negative. But don't go outside of position
6913 * 'lim', which better be < s if off < 0 */
6917 PERL_ARGS_ASSERT_REGHOP3;
6920 while (off-- && s < lim) {
6921 /* XXX could check well-formedness here */
6926 while (off++ && s > lim) {
6928 if (UTF8_IS_CONTINUED(*s)) {
6929 while (s > lim && UTF8_IS_CONTINUATION(*s))
6932 /* XXX could check well-formedness here */
6939 /* there are a bunch of places where we use two reghop3's that should
6940 be replaced with this routine. but since thats not done yet
6941 we ifdef it out - dmq
6944 S_reghop4(U8 *s, I32 off, const U8* llim, const U8* rlim)
6948 PERL_ARGS_ASSERT_REGHOP4;
6951 while (off-- && s < rlim) {
6952 /* XXX could check well-formedness here */
6957 while (off++ && s > llim) {
6959 if (UTF8_IS_CONTINUED(*s)) {
6960 while (s > llim && UTF8_IS_CONTINUATION(*s))
6963 /* XXX could check well-formedness here */
6971 S_reghopmaybe3(U8* s, I32 off, const U8* lim)
6975 PERL_ARGS_ASSERT_REGHOPMAYBE3;
6978 while (off-- && s < lim) {
6979 /* XXX could check well-formedness here */
6986 while (off++ && s > lim) {
6988 if (UTF8_IS_CONTINUED(*s)) {
6989 while (s > lim && UTF8_IS_CONTINUATION(*s))
6992 /* XXX could check well-formedness here */
7001 restore_pos(pTHX_ void *arg)
7004 regexp * const rex = (regexp *)arg;
7005 if (PL_reg_eval_set) {
7006 if (PL_reg_oldsaved) {
7007 rex->subbeg = PL_reg_oldsaved;
7008 rex->sublen = PL_reg_oldsavedlen;
7009 #ifdef PERL_OLD_COPY_ON_WRITE
7010 rex->saved_copy = PL_nrs;
7012 RXp_MATCH_COPIED_on(rex);
7014 PL_reg_magic->mg_len = PL_reg_oldpos;
7015 PL_reg_eval_set = 0;
7016 PL_curpm = PL_reg_oldcurpm;
7021 S_to_utf8_substr(pTHX_ register regexp *prog)
7025 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
7028 if (prog->substrs->data[i].substr
7029 && !prog->substrs->data[i].utf8_substr) {
7030 SV* const sv = newSVsv(prog->substrs->data[i].substr);
7031 prog->substrs->data[i].utf8_substr = sv;
7032 sv_utf8_upgrade(sv);
7033 if (SvVALID(prog->substrs->data[i].substr)) {
7034 if (SvTAIL(prog->substrs->data[i].substr)) {
7035 /* Trim the trailing \n that fbm_compile added last
7037 SvCUR_set(sv, SvCUR(sv) - 1);
7038 /* Whilst this makes the SV technically "invalid" (as its
7039 buffer is no longer followed by "\0") when fbm_compile()
7040 adds the "\n" back, a "\0" is restored. */
7041 fbm_compile(sv, FBMcf_TAIL);
7045 if (prog->substrs->data[i].substr == prog->check_substr)
7046 prog->check_utf8 = sv;
7052 S_to_byte_substr(pTHX_ register regexp *prog)
7057 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
7060 if (prog->substrs->data[i].utf8_substr
7061 && !prog->substrs->data[i].substr) {
7062 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
7063 if (sv_utf8_downgrade(sv, TRUE)) {
7064 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
7065 if (SvTAIL(prog->substrs->data[i].utf8_substr)) {
7066 /* Trim the trailing \n that fbm_compile added last
7068 SvCUR_set(sv, SvCUR(sv) - 1);
7069 fbm_compile(sv, FBMcf_TAIL);
7077 prog->substrs->data[i].substr = sv;
7078 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
7079 prog->check_substr = sv;
7086 * c-indentation-style: bsd
7088 * indent-tabs-mode: t
7091 * ex: set ts=8 sts=4 sw=4 noet: